Professional Documents
Culture Documents
by
Alireza Heidari
Doctor of Philosophy
c Alireza Heidari
August 2015
Sydney, NSW, Australia
PLEASE TYPE
THE UNIVERSITY OF NEW SOUTH WALES
Thesis/Dissertation Sheet
Title: Reliability Analysis of Power Distribution System in Presence of Distributed Generation Units
One of the primary emphases in planning and designing distribution networks is the reliability evaluation considering the economic impact of minimizing the
interruptions for the end users in the networks. An effective way to minimize the customer interruptions is to use protective devices (fuses and CBs) and
sectionalizing switches. In power distribution reliability analysis, the optimal placement of switches and protective devices can reduce the annual failure rate and
outage duration of the customers and consequently improves the reliability indices of the networks. This research presents a new mixed-integer nonlinear
programming model aiming to identify the type, optimum number and location of protective devices and sectionalizing switches in distribution networks more
accurate ly. A value-based reliability optimization formulation is derived from the proposed model. To ensure the effectiveness of the proposed formulation both
technical and economic constraints are considered.
In power distribution reliability analysis, distributed generation (DG units) can potentially improve network reliability by reducing interruption duration and
restoration time after a given fault. However, such improvement depends on the operation mode of the DG units. An island can be formed when sufficient local
generation exists to avoid load prioritization and/or shedding.
The problem of optimum switch and protective device placement becomes more complicated when DG units are present in distribution networks. A method for
solving the problem of optimal placement of sectionali zing switches with high penetration of DG is also presented in this research. The primary objective is to
minimize the total cost of achieving a certain level of reliability in presence of DG. The problem is formulated using mixed-integer linear programming to
determine the optimal number and location of sectionalizing switches.
The accuracy of the reliability analysis can be affected directly by the customer interruption cost model. A probability distribution cost model (PDM) is
developed based on a cascade correlation neural network to have a more accurate reliability assessment. The performances of the proposed approaches are
assessed and illustrated by studying on the RBTS test system and a typical real size network. Simulation results verify the capability and accuracy of the proposed
approaches.
I hereby grant to the University of New South Wales or its agents the right to archive and to make available my thesis or dissertation in whole or in
part in the University libraries in all forms of media , now or here after known , subject to the provisions of the Copyright Act 1968. I retain all
property rights , such as patent rights . I also retain the right to use in future works (such as articles or books) all or part of this thesis or dissertation .
I also authorise University Microfilms to use the 350 word abstract of my thesis in Dissertation Abstracts International (this is applicable to doctoral
theses only).
The University recognises that there may be exceptional circumstances requiring restrictions on copying or conditions on use . Requests for
restriction for a period of up to 2 years must be made in writing . Requests for a longer period of restriction may be considered in exceptional
circumstances and require the approval of the Dean of Graduate Research .
'l
COPYRIGHT STATEMENT
'I hereby grant the University of New South Wales or its agents the right to
archive and to make available my thesis or dissertation in whole or part in the
University libraries in all forms of media, now or here after known , subject to the
provisions of the Copyright Act 1968. I retain all proprietary rights, such as patent
rights. I also retain the right to use in future works (such as articles or books) all
or part of this thesis or dissertation .
I also authorise University Microfilms to use the 350 word abstract of my thesis in
Dissertation Abstract International (this is applicable to doctoral theses only).
I have either used no substantial portions of copyright material in my thesis or I
have obtained permission to use copyright material; where permission has not
been granted I have applied/will apply for a partial restriction of the digital copy of
my thesis or dissertation.'
AUTHENTICITY STATEMENT
'I certify that the Library deposit digital copy is a direct equivalent of the final
officially approved version of my thesis. No emendation of content has occurred
and if there are any minor variations in formatting, they are the result of the
conversion to digital format.'
' I hereby declare that this submission is my own work and to the best of my knowledge it contains no
materials previously published or written by another person , or substantial proportions of material
which have been accepted for the award of any other degree or diploma at UNSW or any other
educational institution , except where due acknowledgement is made in the thesis. Any contribution
made to the research by others , with whom I have worked at UNSW or elsewhere, is explicitly
acknowledged in the thesis . I also declare that the intellectual content of this thesis is the product of
my own work, except to the extent that assistance from others in the project's design and conception
or in style, presentation and linguistic expression is acknowledged .'
Date ... !/;(.o.6.; .2o.J6... ... ...... ........... ..... ... .... ... ..... .... .. .. ... ... ...... ....... .
Abstract
One of the primary emphases in planning and designing distribution networks is the
reliability evaluation considering the economic impact of minimizing the interruptions for
the end users in the networks. An effective way to minimize the customer interruptions
is to use protective devices (fuses and circuit breakers) and sectionalizing switches. In
power distribution reliability analysis, the optimal placement of switches and protective
devices can reduce the annual failure rate and outage duration of the customers and
consequently improves the reliability indices of the networks. This research presents a new
mixed-integer nonlinear programming model aiming to identify the type, optimum number
and location of protective devices and sectionalizing switches in distribution networks
more accurately. A value-based reliability optimization formulation is derived from the
proposed model. To ensure the effectiveness of the proposed formulation both technical
and economic constraints are considered.
In power distribution reliability analysis, distributed generation (DG units) can poten-
tially improve network reliability by reducing interruption duration and restoration time
after a given fault. However, such improvement depends on the operation mode of the
DG units. An island can be formed when sufficient local generation exists to avoid load
prioritization and/or shedding.
The problem of optimum switch and protective device placement becomes more com-
plicated when DG units are present in distribution networks. A method for solving the
problem of optimal placement of sectionalizing switches with high penetration of DG is
also presented in this research. The primary objective is to minimize the total cost of
achieving a certain level of reliability in presence of DG. The problem is formulated us-
ing mixed-integer linear programming to determine the optimal number and location of
sectionalizing switches.
The accuracy of the reliability analysis can be affected directly by the customer inter-
ruption cost model. A probability distribution cost model (PDM) is developed based on
a cascade correlation neural network to have a more accurate reliability assessment. The
performances of the proposed approaches are assessed and illustrated by studying on the
v
RBTS test system and a typical real size network. Simulation results verify the capability
and accuracy of the proposed approaches.
• Proposing the MINLP and MILP-based methods for solving the problem of optimal
sectionalizing switches and protective devices placement in the presence of DG units.
• Considering the cost of interruption due to the temporary faults in the proposed
formulation.
• Proposing a new approach to detect the location of both DGRESs and fault by
using an adaptive protection scheme based on a radial basis function neural network
(RBFNN) and backtracking algorithm.
• The detailed formulations of the probabilistic based analytical technique for reliabil-
ity analysis of the islands in distribution network.
• Proposing a PDM cost model based on a cascade correlation neural network (CCNN).
• Proposing a value based formulation considering the cost of customer outage and
protective devices.
I extend a special thanks to all my friends we met in Sydney, and will always remember
all the enjoyable moments we shared together.
In the end, I can’t find words to express my appreciation to my soulmate, Goli; thank
you for being a perfect spouse and for all your understanding and endless patience.
vi
Contents
Abstract iv
Acknowledgements vi
1 Introduction 1
1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Power Distribution System Reliability . . . . . . . . . . . . . . . . . . . . . 2
1.3 Research Motivation and Definition of the Research Problem . . . . . . . . 4
1.3.1 Protective Device Placement Optimization . . . . . . . . . . . . . . 4
1.3.2 Island Operation of DG . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3.3 The Potential Impacts of DG on Protective Device Placement Op-
timization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3.4 The effect of Customer Interruption Cost Model on Reliability Anal-
ysis of Distribution Networks . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Objective of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 List of Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5.1 Journal Papers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5.2 Conference Papers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5.3 Papers under Review . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
vii
Contents viii
List of Symbols 36
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3 Theoretical Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.3.1 RBFNNs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.3.2 ROLS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.3.3 Back Tracking Algorithm . . . . . . . . . . . . . . . . . . . . . . . . 45
3.4 Island Formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.4.1 Principle and concept . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.4.2 Concerns of Unintentional Islanding . . . . . . . . . . . . . . . . . . 49
3.4.3 Reliability Indices of the Island . . . . . . . . . . . . . . . . . . . . . 49
3.5 Proposed Sub-Islanding Approach . . . . . . . . . . . . . . . . . . . . . . . 50
3.5.1 Offline Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.5.2 Determining Faulty Line . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.5.3 Identification of DGRES Locations . . . . . . . . . . . . . . . . . . . 57
3.5.4 Sub-Island Formation . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.5.5 Main Protection Coordination Algorithm . . . . . . . . . . . . . . . 58
3.5.6 Reliability analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.6 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.6.1 Sensitivity analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.6.1.1 DGRES Location . . . . . . . . . . . . . . . . . . . . . . . 69
3.6.1.2 DGRES Number . . . . . . . . . . . . . . . . . . . . . . . . 69
3.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
List of Symbols 72
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.2 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
Contents ix
xii
List of Figures xiii
5.1 Possible locations for sectionalizing switches and protective devices in radial
distribution network. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.2 Modified distribution test system (with 38 load points (LP), 51 sectionaliz-
ing switches and 38 fuses). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
5.3 Optimal location of sectionalizing switches of case 4. . . . . . . . . . . . . . 123
5.4 Compare TC in Case 4, 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
5.5 TC, ECOST and SC vs. number of available switches. . . . . . . . . . . . . 124
5.6 Optimal location of sectionalizing switches and fuses of Case 7. . . . . . . . 125
5.7 Optimal location of sectionalizing switches of case 11. . . . . . . . . . . . . 129
5.8 Optimal location of sectionalizing switches by the proposed MINLP of [2]
test system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
7.1 Sample of radial distribution network and fault restoration steps. (a) Fault
occurred in LS3, (b) fault clearance step, (c) fault separation step, and (d)
fault restoration step. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
7.2 Sample of fully automated radial distribution network and fault restoration
steps. (a) Fault occurred in LS3, (b) fault clearance step, (c) fault separation
step, and (d) fault restoration step. . . . . . . . . . . . . . . . . . . . . . . . 168
7.3 Possible locations for sectionalizing switches and protective devices in a
radial distribution network. . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
7.4 Test system 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
7.5 Modified distribution test system (with 38 load points (LP), 51 sectionaliz-
ing switches, 38 fuses and 7 CBs). . . . . . . . . . . . . . . . . . . . . . . . 182
List of Tables
xiv
List of Tables xv
Abbreviation Definition
DG Distributed generation
PDM Probability distribution cost model
AERI Australian energy research institute
RBTS Roy Billinton test system
CB Circuit breaker
MINLP Mixed-integer nonlinear programming
MILP Mixed-integer linear programmingr
Std Standard
DNO Distribution network operator
RBFNN Radial basis function neural network
SAIFI System average interruption frequency index
SAIDI System average interruption duration index
ENS Energy not supplied
ECOST Expected interruption cost
CDF Customer damage function
AAM Aggregate or average cost model
CCNN Cascade correlation neural network
DGRES Distributed generation based on renewable energy source
HLI Hierarchical Level I
HLII Hierarchical Level II
HLIII Hierarchical Level III
CEA Canadian Electrical Association
xvi
Abbreviations xvii
MHC
ATMH
BARON Branch-and-reduce optimization navigator
GAMS General algebraic modeling system
RWTS Reliability worth of the TS
RC Relocation cost
UCR Utility cost of reliability
RBF Radial basis function
MLP Multilayer perceptron
SOM Self-organized map
OLS Orthogonal least squares
SSE Sum of squared errors
POF Probability of operational failure
RTS Reliability test system
FOR Force outage rate
LPIF Load point interruption frequency
RMS Root-mean square
BP Back-propagation
CIC Customer interruption cost
RWDG Reliability worth of introducing DG
DAS Distribution automation system
CCDF Composite customer damage functions
SIC Standard industrial classifcation
TCR Total cost of reliability
RWDAS Reliability worth of utilizing DAS
“Behind every story of success is an even greater story of support.”
Dedicated to my family
To My Love,
Goli
To My Knight,
Yamin
To My Little Angel,
Ava
xix
Chapter 1
Introduction
1.1 Overview
Power utilities goal is to provide reliable and economical electricity to their customers. As
power outages and interruptions to customers have a great economic impact, it is notably
vital to plan and design proper power system especially from the reliability point of view.
At present, introducing the restructured electric power utilities led to deregulated electric
power utilities work as distinct parts such as generation, transmission and distribution
companies. In this regard, all these three service providers are responsible for retaining
the reliability of the overall power system instead of only a single electric utility.
The foremost important function of an electric power system is to satisfy the system
load requirement with a reasonable assurance of continuity and quality of supply. The
capability of the system to provide an adequate supply of electrical energy is usually
designated by the term reliability [1]. System reliability can be improved by reducing
the frequency of occurrence of faults and the repair time by means of various design and
maintenance strategies [3]. Electrical utilities continuously install equipment to satisfy the
growing demand for reliability. It is therefore important to objectively assess the benefits
of installing these devices compared with the cost of the equipment. Distribution systems
are normally equipped with reclosers and sectionalizers, which are used to isolate faulted
parts of the network and recover the power supply from temporary faults. Installing new
1
Chapter 1. Introduction 2
protective devices such as sectionalizing switches or circuit breakers could improve the
reliability of the distribution system, but the extent of improvement should be quantified,
which will help to numerically evaluate the installation of such equipment.
Recently, distributed generation (DG) has appeared and started to play an important
role in the electrical systems. However, integrating DGs in the power systems has many
challenges such as voltage alleviation and regulation and protection coordination. One of
the most important issues in terms of adding DGs into a power system is to enhance the
reliability and reduce cost. New research has predicted that DGs will be a considerable
percentage of all new generation and that they would be about 20% of new generations
being installed. DGs have different types of resources and they use different technologies
to provide energy to the power systems. Implementing of DGs has some positive and
negative effects for both utility and customers. Decreasing loss of power, enhancing the
reliability of the system and reduction of system expansion cost are some of the advantages
of adding DGs into a system. In contrast, voltage control, islanding mode and power
quality issues are some disadvantages of using DGs in the power system. Meanwhile, some
accelerating factors such as economic considerations, environmental concern, power system
deregulation and technological advancements push utilities to increase the number of DG
units in their networks. For instance, DGs will improve the utilitys ability to shave the
peak load on a feeder and help to serve loads during contingencies. Thus, utilizing DGs in
power distribution systems will considerably improve reliability indices of the system and
load points. As the DGs have some particular characteristics that distinguish them from
conventional generating units, they could not be treated neither as conventional generation
nor substation to carry out reliability assessment. However, how much contribution DG
unit can provide to the distribution network reliability is a very significant issue that power
utilities most care about.
The term of reliability has a broad meanning. Reliability is a summary term meaning
stability, proper efficiency and dependability. Entirely, reliability distinguishes the ca-
pability of the network to carry out its determined function, where past experience can
Chapter 1. Introduction 3
Hierarchically, the most convenient approach for dividing the system is to use its main
functional zones, therefore power systems are divided into three approximately indepen-
dent areas of operation as follows: generation, transmission (or bulk power) and distribu-
tion systems. The main aim of a power system is to deliver electric power to its customers
as reliable and as economic as possible.
During the past years, the distribution section of the power systems has received sig-
nificantly minor attention in the reliability point of view compared to the generation and
transmission sections and many utilities have focused to enhance the reliability of their
generation and transmission sections. The reluctance of electric utilities to focus on the
distribution system has changed as the generation and transmission reliability concentrates
on the capacity and loss of load probability in the system, with some regard to the system
components, while distribution reliability considers all forms of engineering: planning, de-
signing and operating. Currently, at present, deregulation is forcing electric utilities to
change their past prospect so under the new era of deregulation, the attention has shifted
to distribution systems to provide a reliable and economical service for customers in the
system.
system [5]. Thus, it is important to note that increasing the reliability of the distribution
system is necessary and many utilities are now focused on enhancing their performance to
gain dominant customer satisfaction.
The need to operate electricity distribution networks with high reliability and efficiency
while minimizing associated costs remains a challenge for utilities and network planners.
The main challenge is that increasing the levels of reliability requires greater capital and
operational expenditure [6]. Optimization is therefore required to find a balance between
reliability and cost.
The frequency and duration of interruptions are important reliability indices associated
with the distribution networks. Thus, a considerable effort has been devoted to reduce the
effects of failures by utilizing protective devices such as circuit breakers (CBs), fuses and
sectionalizing switches. The proper number, location and type of such protective devices
play an important role in the reliability of distribution networks by minimizing the impact
of interruptions. In addition, the optimum number and location of protective devices are
important to minimize capital investment cost and maximize customer benefits [2].
The ever increasing growth of DG is associated with a number of benefits for utilities and
customers alike. These benefits include potential improvements of reliability and power
quality, cost reduction of electricity generation due to avoidance of network losses and
environmental benefits because of carbon emission reduction.
DNOs would benefit from an improvement in the overall security and quality of the power
supply system.
DG challenges existing protection designs while the placement and type of protective
devices affect the reliability indices. DG can potentially improve network reliability, as
it can reduce interruption duration and restoration time after a given fault. However,
such improvement depends on DG systems operating in islanding mode. An island can be
formed when sufficient local generation exists to avoid load prioritization and/or shedding.
Therefore, the problem of optimum switch placement optimization becomes more com-
plicated when DG units are present in the distribution network. It is quite often difficult to
determine the optimal position of sectionalizing switches for supporting the normal power
supply of the islanded area in the presence of faults. The objective of this dissertation is to
propose a mixed-integer linear programming (MILP) based method to solve the problem
of optimal sectionalizing switch placement in the presence of DG. Based on the islanding
Chapter 1. Introduction 7
operation of DG, a formulation to calculate the outage time of the load points inside and
outside of the island is proposed. As the outage time of the customers is very important
to evaluate the cost of reliability, this formulation can provide a precise assessment of
reliability indices. This formulation has not been considered in the switch optimization
problem with MILP formulation yet.
The most common customer-oriented reliability indices such as SAIFI, SAIDI and ENS,
were used in [19]- [20] to assess the optimal number and location of sectionalizing switches
in distribution networks with DG. But these indices are not sufficient to represent the
outage cost of customers. Therefore, in this research work, the expected interruption cost
(ECOST) index is used to associate reliability with customer outage costs since it accounts
for the effects of the interruption durations, failure rate of equipment, load variations,
customer types and customer damage functions. Furthermore, the cost of interruption due
to the temporary faults is considered in the proposed MILP formulation. In summary, the
objective is to report results of finding the optimal number and location of sectionalizing
switches to minimize the outage cost of customers considering sectionalizing switch capital
investment, installation and annual maintenance costs in the proposed formulation.
P
UN
SAIDI system average interruption duration index P i i (1.2)
λi Ni
X
ENS energy not supplied La(i) Ui
(1.3)
Chapter 1. Introduction 8
The accuracy of the reliability worth can be affected directly by the customer interruption
cost model. The load points and system reliability indices can change due to the differ-
ent cost models. A variety of methods have been introduced to evaluate the customer
impacts due to interruptions [21]. These methods can be divided, based on the method-
ological approach, into three extensive groups: analytical evaluation methods, case studies
of blackouts, and customer surveys [22].
The customer survey approach provides customer outage cost data that could be mod-
eled as the customer damage function (CDF) [23]. This model determines the aggregate
or average cost of interruption for each customer sector as a function of duration. Con-
sequently, the basic cost models to implement reliability worth analysis can be derived
by CDF. The main disadvantage of the aggregate or average cost model (AAM) is that
the dispersed nature of data within a specified customer group is neglected. The dis-
persion of the customer interruption cost data is important information and can have a
significant effect on the accuracy of the reliability assessment. The dispersed nature of
the customer interruption costs at specified failure duration has been designated as the
probability distribution cost model (PDM) [24].
The research work described in this dissertation examines the effect of customer inter-
ruption cost model on distribution system reliability indices. A cascade correlation neural
network (CCNN) is used to develop two interruption cost models including AAM and
PDM. A contingency based analytical technique is adopted for reliability worth analysis.
In addition, the impact of introducing distributed generation units into the network is
presented.
• An MINLP formulation to solve the optimal type, number and location of section-
alizing switches and protective devices problem, simultaneously.
The following publications have been developed during the course of the Ph.D. studies.
The following papers have been published or accepted for publication in peer review jour-
nals:
[4] A. Heidari, V. G. Agelidis, and M. Naderi, “Effects of switch type and location on the
reliability of power distribution systems considering distributed generation,” in Proc.
IEEE International Conference on Industrial Technology (ICIT), South Africa, Feb.
2013, pp. 710–715.
[5] A. Heidari, and V. G. Agelidis, “Switch designation and distributed generation im-
pact on reliability of power distribution systems: Worth analysis,” in Proc. IEEE
TENCON Spring International Conference, Australia, 2013, pp. 352-357.
The following papers have been submitted to peer review journals and are currently under
review:
[13] A. Heidari, V. G. Agelidis, and J. Pou, “Reliability Cost/ Worth Analysis of Distribu-
tion Systems With Distributed Generation Considering the Customer Cost Model,”
IEEE System Journal., (First revision).
These journal and conference papers are associated with the chapters of the thesis as
shown:
Chapter 1. Introduction 12
Chapters Publications
Chapter 3. Distribution System Reliability Analysis
[2], [8] and [10]
Considering Distributed Generation
Chapter 4. Distribution System Planning Considering
[1], [3] to [5] and [9]
Optimal Placement of Switching Devices
Chapter 5. Optimal Type, Number and Location of Protective
[9] and [12]
Devices in Distribution Systems
Chapter 6. Reliability Analysis of Distribution System
[6] and [13]
Considering Customer Cost Model
Chapter 7. Distribution System Protective Devices Placement
[6] and [12] to [14]
Optimization Considering DAS and PDM
• Chapter 6 introduces the principle of the CCNN. Then the CCNN is used to construct
the AAM and PDM cost models for reliability worth assessment in distribution
networks. A contingency-based analytical technique is adopted for the reliability
analysis. The details of the reliability worth assessment are discussed in this chapter
through numerical examples.
• Finally, Chapter 7 concludes this dissertation and provides discussions and further
suggestions on possible future studies.
Chapter 2
2.1 Introduction
The definition of reliability in the literature address some common aspects of power systems
such as meeting customer demands, continuity of service, and the vulnerability of the power
system. The concept of reliability in the power system can be construed by the following
three different aspects:
1. adequacy, as the ability of the power systems to supply the aggregate power and en-
ergy requirement of all customers in the system at all times taking into consideration
scheduled and expected unscheduled outage of the system components [25];
2. security, as the ability of the power systems to withstand sudden disturbances such
as unexpected loss of system components or electric short circuits [25];
3. quality, as the measure, analysis, and improvement of the bus voltage to meet the
voltage and frequency limits [26].
In fact, the reliability analysis cannot be considered apart from the objectives of the
system. The goals of the distribution system may be discussed as (1) covering the territory
(an aspect of adequacy), (2) meet the peak demand (another aspect of adequacy), (3) the
ability of operating under adverse conditions (security), and (4) providing a stable voltage
14
Chapter 2. Distribution System Reliability 15
Reliability
Generation
- Stability - Frequency
- Security - Duration
Reserve Margin - Component - Timing (hour,
reliability season, etc.)
- Spinning & hot-start - Magnitude (partial or
total)
Distribution reserves
- Advance notice
- Area control
- Harmonics - Persistence
Frequency & Duration - Under freq. load - Coverage
of interruptions shedding
- Automatic load and
generation rejection
Fuel supply
Adequacy of energy supply
capability
and current (quality). Therefore, the goals of the distribution system are compliant with
the reliability concerns.
Note that, different perspectives may have different definitions for the reliability. The
customer and utility perspectives are the two main perspectives for reliability consideration
of a power system. From the customer point of view, the reliability concerns depend on
customer end-use patterns. In fact, customers associate service reliability with restoration
time and how accessible and responsive the utilities are during interruptions or outages.
Therefore any outage in service is undesirable from the customer perspective. The utility
perspective is different from the customer perspective. The definition of reliability for
the utility should be related to both service reliability on the service side of customer
load point and reliability of the supply side which includes the reliability of generation,
transmission and distribution assets [27]. Fig. 2.1 provides a summary of the reliability
concern from different perspectives [27].
Chapter 2. Distribution System Reliability 16
Three traditional hierarchical levels of the power systems are shown in Fig. 2.2 [7]. Hier-
archical Level I (HLI) just includes the generation system and from the reliability point of
view it is necessary to consider and determine generation capacity in order to ensure the
desire demand. The reliability evaluation of hierarchical Level II (HLII) comprises both
generation and transmission networks, which is referred to as the bulk power network or
composite systems. To certify satisfactory energy transfer levels from generation units to
bulk customers, the transmission system should be implemented [28]. Hierarchical Level
III (HLIII) contains all the three generation, transmission and distribution systems. Be-
cause of the significant amount of problems in this level, the overall reliability assessment is
rarely carried out. In contrast, the reliability evaluation of generation systems, composite
generation and transmission systems and distribution systems are conducted individually.
References [29, 30] performed comprehensive reliability evaluation considering all part of
the power systems.
based on this general approach are not precise and only provide an upbeat view of the
system performance.
Some analysts [32–34] have attempted to meliorate this issue by using λ which is
defined as the average failure rate and U which is defined as the annual average unavail-
ability for the equivalent transmission system components that are derived basically from
an individual composite power system reliability assessment, as the input for evaluating
distribution system reliability. The main disadvantage of this approach is that its a bit
difficult to recognize the root outage cause location within the transmission system.
that the distribution system makes the most individual contribution to the overall customer
supply unavailability. They have demonstrated that the bulk power network contributes
just a small part to the overall Hierarchical Level III (HLIII) load point indices, which
were obtained from historical events reporting [7]. Accordingly, this research will focus on
evaluating distribution system reliability and how to improve it.
Basically, in order to prove the accuracy and validity of the results, many methods are used
in the reliability evaluation of power systems (Fig. 2.3). These methods are divided into two
major approaches, simulation and analytical point of views. These approaches have some
limitations and competencies. In this subsection, the main concepts of some commonly
used methods in these two techniques are reviewed and the summary is presented in
Table 2.1.
All the analytical approaches use mathematical models to evaluate the reliability of power
systems. First of all, the system topology and input values are determined and then
a mathematical model of the power systems is created. Finally, calculation results are
obtained for the system. Some of the commonly used methods are block diagram, event
tree, cut set, fault tree, state enumeration, Markov modeling and failure effect analysis.
Although using reliability sets in the calculation has appeared in the recent years, the
fact of making assumptions and using approximation are still major problems of these
approaches.
This method detracts the whole parts of a power system to a simple equivalent network
with just series and parallel components through utilizing modular concepts. then, a
qualified probability approach is used to compute the reliability indices for the system with
Reliability Modeling
Simulation Analytic
Assessment
Techniques for
Load Point Component
Outage Exposure
Outage Failure
Chapter 2. Distribution System Reliability
this series and parallel alignment [35–40]. In this method, in order to convert the system
into its functional model, it is necessary to oversimplify component reliability indices,
which is the main disadvantages of this method. Reference [41] performed estimation for
failure rate values according to the number of pieces of equipment that belong to each
functionally group such as control, monitoring and protection groups.
An event tree of a system is defined as a visual display of all events that may occur in a
power system [42, 43]. The application of this method needs to run an event, and then
perceive the possible consequences covering success and failure of the system components
as the main branches of the tree. A failure path demonstrates a failure scenario that the
network is nominated to fail if all components in that specific path fail. The debility of
this method is that the size of the tree can be astounding in the case of large networks. For
instance, if each component has both operating and failed status, a complete event tree
for the system with n components has 2n paths. Therefore, many authors use a reduced
event tree that decreases the accuracy of the results.
A set of system components whose failure causes the failure of the power system are
described as a cut set. Minimal cut set is a concept that refers to the minimum subset of a
cut set, which contains a set of components in the system that their failure leads the system
to fail. All cuts in the cut set are in series with each other, but the components inside a
cut are incorporated using the parallel components principle. In this method, the minimal
cut set can be derived in order to decrease the size of the tree mentioned in the previous
subsection. The failure path of an event is referred to a cut in the cut set. However,
minimal cut set is also excessively large for huge networks. Basically, approximations such
as neglecting cut sets larger than a certain value are neglected in order to achieve faster
response from the calculations. This is derived from the assumption mentioned that high
order cut sets are much less probable than low order ones [42]. However, an exhaustive
assessment for accurate results is missed in this method.
Chapter 2. Distribution System Reliability 21
A fault tree is a figurative representation of the failure incorporated in the system. Unlike
the event tree, the top event of the tree can only be a specific failure situation. Application
of this method needs to construct the branch events as the required events in order to lead
to the top failure event. Accordingly, this method is applied for safety evaluation as
a mission oriented assessment. However, this method is not used for precise reliability
evaluations [44, 45].
The concept of this method is finding the events that have an adverse effect on power
system reliability and then assess their effects. In fact, considering all the possible contin-
gencies is not necessary and sometimes is not practical. In this method, the contingency
screening policy is used to decrease the number of estimated states. Primary contingencies,
whose outage frequencies exceed from the initial value, are considered in the assessment
as an important approach. Also, there is another approach that uses some accuracy situa-
tion such as overload condition of specific contingency. In these approaches a contingency
ranking analysis is used [35, 46, 47]. The main limitation of this method is that in case of a
large system the number of contingency is too large, and using the screening method is not
sufficient [48, 49]. Also, it is difficult to incorporate the stochastic behavior of the system
loading. Generally, just a few load levels based on past experience are utilized to carry
out the analysis. Furthermore, considering planned outages is not easy to be incorporated
into this method.
The Markov modeling method uses a matrix that is appropriate for reliability evaluation of
memory-less systems whose probability distributions of components are fixed rates [50–53].
In this method, a stochastic transitional probability matrix is created in order to perform
a reliability evaluation on the system. The dimensions of this matrix in the case of a
system with n state are n×n square. Since the system topology changes, it is necessary
Chapter 2. Distribution System Reliability 22
to re-cerate the system transitional matrix. Thus, this method is generally suitable for
systems with simple configuration.
In [54], eight reliability sets are proposed for calculating SAIDI of particular feeders or
circuits. Basically, these sets are determined based on the segment of the feeders. Imple-
mentation of this method requires simplification of the system model to sets that consider
an average failure rate value for each segment. Then, an optimal DG placement strategy
with the presence of time varying loads in the system is assessed. The main limitation
of this method is that it is only suitable for radial systems. Besides, this method cannot
consider multiple failures, as it is assumed that only one failure takes place at a time.
Failure mode and effect analysis (FMEA) is a systematic technique for failure analysis. It
involves reviewing the components of the system to identify failure modes and their effects.
This method is used to identify the potential failure modes of each functional component
of the system and study the effects that these failures might have on the system. FMEA
can be used for maintenance planning in the system. It can be simply qualitative, or it can
include the quantitative assessment of different failure as well. The failure models depict
how the system components can fail, how often they might fail and how long it takes to
repair them when they fail [50, 51, 54–60].
The FMEA takes the power system through a number of post-fault operational states
that can include:
The objective of the FMEA function is to determine if the system faults will lead
to load interruptions and if so, which loads will be interrupted and for how long. The
procedure of evaluating the system response after a specific contingency is as follows.
• Fault Clearance;
• Fault Isolation;
• Power Restoration;
• Overload Alleviation;
• Load Transfer;
• Load Shedding;
Note that, a survey can be conducted to appraise the customer interruption statistics
for different customer sectors. The survey data have been assessed to extract the sector
customer damage function related to each group of customers.
The present analytical reliability techniques are not able to model large power distribu-
tion systems and their components for reliability evaluation. The methods introduced
previously in this section use parallel and series network calculation principles. One of
the limitations of those analytical methods is that there are many real systems that are
not easy to separate them into small independent subsystems [61–64]. Another problem
with these approaches is the difficulty in the calculation when using traditional matrix
methods. Furthermore, in the case of using the Markov method, as described before, the
events are memory-less. The simulation or Monte Carlo approach is more universal than
the analytical approaches. This method provides a solution for complex problems in the
Chapter 2. Distribution System Reliability 24
system that is not possible to solve with analytical methods. It is also, used to simu-
late the probabilistic behavior of networks. The random generator of the Monte Carlo
simulation is used to make random variants that track the distribution functions, even
non-constant venture rates. In this approach, the simulation convergence takes place as a
fluctuation process when the estimated result become closer to the true value as a sam-
ple size increase [51]. Also, a factor of variation of the outputs in the system is used as
a convergence criterion. The Monte Carlo method is divided into two main approaches
which are sequential and non-sequential approaches. The sequential approach is used for
simulating the chronological status of the system operation. This approach is also called
state duration sampling method [65]. In this method, in order to perform the chronologi-
cal sequence of the system state it is necessary to generate the chronological sequence of
states for each component in the system and combine them together. This method is able
to evaluate the complexity failure events that sometimes are integrated into the reliability
models. The evaluations are also include time varying component failure rates owning
to weather conditions or previous system states. In order to assess cascading outage in
the analysis of the system, utilizing the sequential method is required, as the subsequence
events in the cascading outages take place sequentially [66, 67]. Reliability indices for
the system are computable by using power system analysis methods, such as contingency
analysis. With this method, we are capable of using historical hourly weather records to
choose proper components failure rate in the system. In contrast, the non-sequential ap-
proach is performed by ignoring the event chronologies in the system. The state sampling
method is used to sample the state of each element in the system, and then incorporate
them to define the system state. In this approach, in order to calculate reliability indices,
the system state transition is used instead of component states. Thus, the non-sequential
method cannot predict the interruption duration for the system failure condition and re-
flect the situation of previous contingency cases the loading of remaining feeders in the
stressed systems. The evidences from previous studies suggest that, the sequential method
is preferred when intricate operating conditions are applied.
The main weakness of simulation methods is that to obtain an acceptable level of the
accuracy of the performance indices, a huge amount of experiments need to run. Due to this
long computational time, simulation approaches are not popular as analytical approaches.
Table 2.1: Reliability Methods Summary
In the previous sections, methods to evaluate reliability indices in the power distribution
systems have been introduced. The significant of using such predictive methods to assess
reliability of the system is to distinguish required procedures to enhance power system
efficiency. In order to choose the best alternative, we should consider different measures
like upgrading and maintenance of the systems. In this section, some methods to meliorate
the reliability of the system are depicted. However, the methods explained here are not
perfect, they are described to give a notion of some of the process taken to improve the
reliability indices of the systems.
2.4.1 Automation
Nowadays, distribution automation has become a top priority for utilities. Distribution
automation is defined as the complete automation of all controllable equipment and func-
tions in the distribution networks (Fig. 2.4). The main tasks are to increase distribution
reliability, improve distribution operations and maintenance, faster the disturbance anal-
ysis and fault location, asset monitor for aging infrastructure and avoid asset overload,
and increase distribution power quality. In recent years, newer applications such as fault
detection, fault location analysis, volt and var control and power quality measurements
have appeared by using distribution automation, all these issues can greatly improve the
reliability of the networks. Also, the application of distribution automation is predicted
to increase in the future years (Fig. 2.5).
One of the major concerns for the electricity supply systems is to provide high supply
reliability to the customers, hence the utilities are turning their consideration to enhancing
efficiency and control in the distribution part of the system which lies between the substa-
tion and the meter. As mentioned before, the reliability of distribution systems is affected
by a number of factors and typically measured by indices such as the SAIDI and SAIFI. In
general, the power quality performance of power distribution systems is steadily focused
on the system operators. In this regard, cost-effective analysis and concepts for system
Chapter 2. Distribution System Reliability 27
development and operation are necessary when automation is used in the networks. More-
over, from the economic system operation point of view, the explicit performance targets
demanded by regulators and customers are becoming more and more important factors.
Several types of protective devices are studied for reliability evaluation in [71]. The
authors quantified that a reliability improvement can be achieved by using each of these
devices, as well as the combination of them. They used a reliability analysis program
designed to determine the methods for improving reliability of the distribution systems.
Reference [72] discusses the challenges and strategies for promoting reliability of the sys-
tem when reclosers, sectionalisers and other switching devices are used in distribution
feeders. More recently, some technical papers have appeared that offer optimum alloca-
tion of switching devices in distribution systems [72]. Automated restoration to improve
reliability is studied in [73]. The analysis of when to restore and how to restore the service
has been proposed in [73], through the use of an offline calculated table and properties
of the binary bus connection matrix are required. A reliable communications system, one
that is open, standards-based, scalable and secure is required in order to gain the benefits
of distribution automation, which the reliability improvement is one of the major ones.
2.4.2 Maintenance
Maintenance experts agree that replacing a component before encountering a failure (pre-
ventively) could, under certain conditions, make better economic sense than replacing the
Chapter 2. Distribution System Reliability 29
As mentioned before, most of the interruptions are due to equipment failure. The key
factor is to distinguish whether the preventive replacement of a specific component is rea-
sonable and, if so, to determine the best time to replace the component. In order to know
when and what actions are appropriate to prevent failure in the system two requirements
must be met. First, as the components become older, they are more susceptible to failure
and from the reliability point of view this means that the components have an increasing
failure rate. In other words, preventive maintenance can help when the component gets
worse with time. The second requirement is that the cost of preventive maintenance should
be less than the cost of corrective maintenance. If both of these requirements are satis-
fied, then preventive maintenance is appropriate and the optimum time (minimum cost) at
which the preventive replacement should take place can be calculated. In this regards, util-
ities have to increase inspection and monitoring of the components, as this process provide
adequate information about what is prone to fails. Equipment conditions data sheet such
as testing insulating oil in transformers and switches, counter measurements of disconnect
switches, and protection devices etc. give an intend of checking the equipment activity
and predict future actions with reasonable precision by failure probability and what and
when maintenance is necessary to improve actuation of equipment. The presence of trees
and vegetation growth around the conductors in the system increases the fault occurrence
and thus failure probability of the system. The management of vegetation is a decretive
act and takes a huge part of utilities resources. Also, vegetation growth leads to increase
losses in the system, fires and physical damages to infrastructure in the system.
Another important method that can be helpful to improve efficiency of the equipment
is load management which helps the system to operate better in normal situation and can
reduce losses and enhance the performance of equipment [74, 75]. Utilities use different
scenarios to implement maintenance procedures and monitor different types of equipment
in the system. Monitoring and maintenance of equipment detract their failure rate and
Chapter 2. Distribution System Reliability 30
Operating Time
Figure 2.6: Cost per operating time unit vs. operating time [1].
afterwards the interruption frequency and duration of failure that their customer may
experience. Some of the procedures that are used by utilities for maintenance are:
• Corrective Maintenances
• Preventive Maintenances
DG mentions to stationary usages of electric generating technologies which are smaller than
15MW of net generating capacity. They may be owned by electric utilities, commercial,
industrial or residential energy consumers, or by independent energy producers. They also
include generating technologies such as diesel engines, small and micro gas turbines, fuel
cells , wind turbines and solar PV and may be combined with electric storage technologies
Chapter 2. Distribution System Reliability 31
such as batteries and flywheels. The capacity of these resources is growing rapidly in
the world (Fig. 2.7) and they are typically installed for one or more of the following
applications:
• Peak shaving: using on-site generation intermittently can be applied for a limited
number of hours per year to shave the peak power demand. From the end users point
of view DG is useful to avoid purchasing grid electricity during expensive peak-rates.
Considering the commercial time of use and critical peak pricing from most utilities,
customers will be able to save a significant part of their energy bill by controlling
their peak demand. The most benefit is provided for the customers with high daily
peaks such as office buildings and retail stores with nightly shut-downs. For instance
Fig. 2.8 shows the curve that represents power purchased from the utility during
the day. When the consumption goes over a particular threshold (700kW), the peak
price of the electricity is much higher. To avoids this high charge, DG can be used
prior to reaching the high price threshold. For getting the best performance, it is
better to run DG at full power and the remaining power be purchased from the
utility. When the load is less than the peak price threshold, the DG turns off and
all the power is purchased from the utility (Fig. 2.8);
Chapter 2. Distribution System Reliability 32
400
300
DG capacity
• Grid support: From the utilities point of view, they install DG companies to sup-
port the power systems for a wide variety of reasons, including meeting highest peak
loads without building new infrastructure; serving remote of off-grid loads; main-
taining power quality to reasonable levels; postponing and sometimes alleviating the
need to upgrades systems; maintaining uninterruptible power during planned outages
to ensure that sensitive loads receive premium quality electricity, and; supporting
new, unplanned load growth;
Alternatives to assess the effects of inetgrating DGs to the power distribution system
are overviewed in this section. Using some renewable energy sources such as wind and
solar, and considering the fluctuating nature of these resources to analyse reliability in the
distribution system are presented in [74]. The developed generation system is modeled in
this work. The authors determined the conventional fuel source units to one subsystem
then the effects of renewable energy are assessed by modifying the model of the system. A
combination of system models is considered to calculate the frequency of capacity shortage
and expectation for loss of loads, which each subsystem plays a role as a multi-state unit.
In order to generate hourly loss of load expectation, they combine the subsystems by
using the cumulative method, where a discrete state algorithm is applied. There was a
significant difference in their results with the traditional methods when a high penetration
Chapter 2. Distribution System Reliability 33
of the fluctuating units is used. Reference [74] follows that method and extends it to assess
the loss of load expectation and unsupplied energy and also enhances the computational
output. In the developed method, the authors used the clustering algorithm to associate
the hourly load to the output of the distributed generation units and dedicate them to
the particular states. The mean values for the loads, probability of each cluster and the
output of distribution generation units are taken by dividing the number of cluster over the
total number of observations. In order to evaluate the reliability indices in the system, the
authors combined all the subsystems and then calculated the indices for each cluster. The
validation of their proposed approach is supported by a case study, which they reduced
the computing time by using clustering approach.
The time sequential simulation is used in [75] to assess the reliability of the power
distribution system in presence of wind turbine generators (WTGs). In this study, the
output power of the wind generator at a determined hour is declared as a function of the
wind speed at that hour and the rated power of the WTG. In order to investigate the
common effects of both the force outage and wind speed the authors in this work made
a multi-state (six-state) model for the wind turbine generator. A two-state model is used
to model other component in the power distribution system such as transformers, etc.
The system topology and switching placement are the most important issues to assess the
reliability of the system. The reliability indices for each load point depend on the location
of that load point in the system. Also, the reliability of the system depends on the other
parameters of the WTG such as cut-in speed, cut-of speed and the average wind speed.
The authors have shown that by choosing the optimal number of wind turbine generator
units the reliability of the power distribution system could have a significant improvement.
DG has positive and negative impact on the power distribution systems, some of the
positive impacts are reactive power compensation, loss reduction, regulation, load power
consumption tracking, spinning reserve and enhancement in the reliability by using the
DGs as a backup sources [76, 77] wherein the DG can provide extra power generation
and supply more customers in the system . To analyse the application of DGs in the
distribution systems a commercial software package is developed in [76]. In this study,
DGs work as backup sources with a voltage source role and are modeled as a transferable
Chapter 2. Distribution System Reliability 34
switches. Another scheme in this paper was used to model the DGs, which DGs as a nega-
tive load inject, power into the system independently of the system voltage. According to
this strategy, DGs could be reduced the load transfer requirements when a fault occurred
in the system, so the reliability of the system could be enhanced. In order to show the
validity of this method, the authors used a test system consisting of five feeders which one
of with a DG unit and four other feeders are connected to the first one via normally open
tie switches. The simulation results have proved significant improvement in the reliability
indices for a certain capacity of DG. However, increasing the capacity of the distributed
generation from a specified value causes the back feed problem and led to abatement of
the power distribution system reliability. Authors in [78, 79] suggested an optimal method
for sizing and siting the DGs into the distribution system, which helps to enhance the reli-
ability of the system. This optimal procedure is determined by implementing a sensitivity
analysis for the power flow equations. Load condition problems in the system are solved
as a security constrained optimization issue and size selecting method is concentrated to
the generation penetration levels. In addition, in that distribution system with DG units,
a genetic algorithm is used to distinguish the optimal place for siting the recloser into the
system. In this study, some reliability indices for the system such as SAIFI and SAIDI are
also developed in order to minimize the objective function for solving the optimal recloser
placement problem. Also, the reliability indices are computed for each zone of the test
system. The results illustrated great reduction in the reliability indices based on using
the optimal strategy for siting reclosers in the distribution system. A probabilistic model
for the reliability analysis is presented in [78] where, inserting DG into the distribution
system improved reliability through balancing demand for the customer with lower rate.
The authors determined an equivalent model for the DG with analogous reliability and
load requirements. According to this study, to avoid making additional feeders with high
capital cost, DG installation in that area is a reasonable solution which, also help to con-
trol the voltage of the system. In reference [80], a comparative analysis is performed to
assess the reliability of the system. Also a commercial software package is used and some
reliability indices such as SAIFI, SAIDI, ENS and ECOST are calculated and compared
in two approaches.
Chapter 3
35
List of Symbols
Symbol Definition
SI
rm Average repair time of the mth sub-island
i
rm Average repair time of component i the mth sub-island
Ljkm Average load of the kth type customer located at the jth load point of
mth sub-island
CDFijkm Customer damage function
di (t) Desired value in output layer
(t) Error term
λSI
m Failure rate of the mth sub-island
λLPi SI Failure rate of LPi in the sub-island
λLP i,DGRES Failure rate of LPi in the presence of DGRES
λLPi Failure rate of LPi without DGRES
λim Failure rate of component i the mth sub-island
If Fault current
Pf Forced outage rate of a DGRES
Φ(ν) Gaussian function
Xk Input vector of RBFNN
Θ Matrix parameter
If Maximum fault current
Wji Neuron weight
nI Number of input layer
nh Number of hidden layer
no Number of output layer
36
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 37
ns Number of source
fi (X) Output function of RBFNN
ULPi Outage time of LPi without DGRES
ULPi SI Outage time of LPi in the sub-island
ULP i,DGRES Outage time of LPi in the presence of DGRES
If Phase current
Ia phase A current
Ib phase B current
Ic phase C current
Ag Phase A to ground fault
Bg Phase B to ground fault
Cg Phase C to ground fault
AB Phase A to phase B fault
AC Phase A to phase c fault
BC Phase B to phase C fault
ABg Phase A and B to ground fault
ACg Phase A and C to ground fault
BCg Phase B and C to ground fault
PIP Probability of island being successful
PGDGRES Probability of a DGRES generating power
PGLP Probability of the load having a certain value
Cj RBFNN center
σ Real constant
Pj (t) Regressor
Φ Regression matrix
η Regularization parameter
tr Repair time
ABC Three phase fault
NSI Total number of sub-islands
Ncm Total number of component in the mth sub-island
NLP Total number of load points
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 38
3.1 Introduction
The ever increasing growth of DG is associated with a number of benefits for utilities and
customers alike. These benefits include potential improvements of reliability and power
quality, cost reduction of electricity generation due to avoidance of network losses and
environmental ones thank to carbon emissions reduction.
of an unintentional island and cease to energize the connected system within two seconds
of the island forming. However, disconnecting DG units when the utility supply is lost
reduces the benefits associated with high penetration of DG. The tripping scenario is
inefficient and leads to unnecessary loss of power from the DG units. An alternative is
to preserve connections to the DG units during system disturbances and intentionally
operate an island to enable continuity of supply and potentially improve the reliability of
the power distribution system [16]. Such an approach would allow several benefits of DG
in islanding operation to be realized by both customers and distribution network operators
(DNOs). In particular, DNOs would benefit from an improvement in the overall security
and quality of the power supply system.
A significant amount of technical work has been published in recently on the intentional
islanding operation of DG. In particular, island detection techniques based on (i) remote
[81], (ii) passive [82], (iii) active [83], (iv) hybrid [84] and (v) grid-reconnection detection
[85] have been explored and tested. In [86], an adaptive intentional islanding configuration
based on the load states and generation availability in the system was proposed. A sam-
pling approach to control the islanding operation state or synchronizing operation state
of the system was demonstrated in [87]. In [88], an impact study was conducted on inten-
tional islanding of DG connected to a radial sub-transmission system. Through a dynamic
study, the authors showed how to manage DG and electric power systems for intentional
islanding in specific cases. Hourly generation models for each type of renewable DG unit
based on islanding operation mode was explored in [89].
the main concern. In each of these studies, the reliability of the distribution system is not
considered when DG units operate in an island mode.
This section presents a background on RBFNN theory and its associated learning method.
The principles associated with the backtracking algorithm used in the proposed sub-
islanding procedure are also presented.
3.3.1 RBFNNs
The RBFNN is a feed-forward neural network which has a topology of three layers, namely,
an input layer with nI dimension, a hidden layer with nh dimension and an output layer
with nO dimension. In the input layer there is one neuron and the neurons for each
predictor variable feed the values to each of the neurons located in the hidden layer.
The hidden layer has a variable number of neurons where, each neuron consists of radial
basis function centered on a specific point with as many dimensions as for the predictor
variables. Each neuron is weighted (wji ) and multiplied by the output value associated
with the neuron in the hidden layer and passed to the summation to create the output
of the network. Fig. 3.1 presents a generic architecture of the RBFNN. Three different
layers of the RBFNN are presented through this figure. The output of the network with
mapping of fi : RnI → RnO , is a linear activation function defined by:
nh
X
fi (X) = W0 + Wji Φj (k X − Cj k) (3.1)
j=1
where X∈RnI is an input vector, k·k denotes the Euclidean norm, Wji are the weights,
Cj ∈RnI , 1 ≤ j ≤ nO , are recognized as the RBFNN centers, and nh is the number of
centers. Furthermore, in the case of multi-output RBFNN, the analysis has been done by
separating the network into a group of single output RBFNNs.
Theoretical studies and empirical results indicated that the nonlinear Φ(·) does not
have a crucial impact on the RBFNN precise form and its performance. For this reason,
in this study the Gaussian function is chosen as the radial basis function and it is given
by:
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 43
−ν 2
Φ(ν) = exp( ) (3.2)
σ2
where σ is a real constant, Φ(ν) → 0 to Φ(ν) → ∞, and implies to the radius of the cluster
represented by a center node.
is how to choose appropriate centers from the data set. In this dissertation, a conventional
method is implemented choosing arbitrarily some data points as centers. This method
cannot guarantee suitable efficiency regarding the requirement that the centers need in
order to appropriately sample the input domain. To select the centers, a regularized
orthogonal least square (ROLS) learning procedure [17] is used. Accordingly, the adequate
and parsimonious RBFNNs can be acquired.
M
X
di (t) = Wji Pj (t) + i (t) (3.3)
j=1
where di (t) is the desired value in the output layer, and the Pj (t) are recognized as the
regressors which are fixed functions of X(t):
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 44
3.3.2 ROLS
The main duty of network learning is defined to select the appropriate centers Cj and
define the corresponding weights Wji . The full mode is defined by considering a training
set of N samples {d(t), X(t)}N
t=1 , where d(t) is the network output corresponding to the
input X(t). All of the training data {X(1), ..., X(N )} are considered as candidates for
centers. We can express (3.3) in the matrix form:
D = ΦΘ + E (3.5)
The least square principle can be used to solve the matrix parameter Θ. An orthogonal
decomposition of the regression matrix Φ can be presented as follow :
Φ = WA (3.6)
D = WG + E (3.7)
where
is defined as an orthogonal weight vector. The classical and modified Gram-Schmidt [94]
can be used to derive (3.8). Afterwards, (3.8) is used to solve Θ. The evidence procedure is
affiliated by the ROLS algorithm in order to estimate a regularization parameter. Likewise,
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 45
a subset of considerable regressor based on the following regularized error criterion is chosen
by ROLS [17].
γ ET E
η= . T (3.10)
N −γ G G
where
n
X WkT Wk
γ= (3.11)
k=1
WkT Wk + η
γ is denoted as the number of good parameter measurements [94] and n is the size of the
subset model.
The different search algorithms developed to solve constraint satisfaction and optimization
problems can be divided into two categories: systematic and opportunistic search methods
[95]. The backtracking search algorithm is based on heuristics and is an example of a
systematic search method which guarantees to find a consistent solution given enough
time. This method is commonly used in artificial intelligence (AI) systems satisfied with
certain constraint conditions.
Figure 3.2: The backtracking search algorithm in distribution network with DGRESs
satisfied by the backtracking condition [96]. Accordingly, finding the solution by moving
forward is done by implementing the backtracking algorithm. This method has been used
to solve syntactic problems [97] by improving the search solution.
Island formation occurs when a part of the distribution system that includes one or more
DGRESs becomes electrically isolated from the rest of the system after a fault clearance
procedure but continues to be energized by the DGRESs in order to supply the loads of
the healthy part of the system. An island is successful only when there is sufficient local
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 47
110kV
220kV
CB6
Island
Sub-Island 3
11kV
11kV DGRES
CB3
CB4
CB2
11kV CB1 11kV CB5
11kV
Sub-Island 4
generation available to supply loads within the islanded zone. If the local generation is
insufficient, load shedding is required to prevent malfunction of the system. Depending on
the number and location of the DGRESs connected to the distribution system, a number of
sub-islands can be formed or wide island coverage can be achieved. An example of multiple
sub-islands and different islanding coverage is presented in Fig. 3.3. These islanding and
sub-islanding zones can be formed by opening the associated protective devices dedicated
to configure the island or sub-island. Note that, Based on the location of the protective
devices, especially circuit breakers and DGs, the utilities will make decision on where
intended island need to be performed.
• Personal safety
• Power quality
• Earthing or grounding
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 48
• Protection
revenue associated with the extensive improvement in the quality and reliability of their
services. As from the customers point of view, the reduction in the frequency and the
duration of interruptions are the main benefits of sub-islanded DGRES operation.
• power quality
• out-of-synchronism closure
• grounding or earthing
• personal safety
• protection coordination
Islanding operation of DGRESs is a feasible solution when all of the above issues
are resolved. The solutions and requirements for successful islanding operation are well
summarized and explained in [100].
X
PIP = ( (PGDGRES × PGLP )) × (1 − Pf ) (3.12)
where PGDGRES represents the cumulative probability of the DGRESs generating power
equal to or greater than certain level, PGLP represents the probability of the load having
a certain value and Pf represents the DGRESs forced outage rate.
In this chapter it is assumed that the load points in the sub-island will be supplied
continuously when the sub-island is forming. Considering that the load point LPi is in the
sub-islanded area with the probability of PIP , the failure rate and associated outage time
of LPi inside the sub-island can be calculated as follows:
where λLPi SI and ULPi SI are failure rate and outage time of LPi in the sub-islanded zone,
respectively. λLPi and ULPi are failure rate and outage time of LPi without DGRESs,
respectively, also λLP i,DGRES and ULP i,DGRES imply on failure rate and outage time of
LPi in the presence of DGRESs, respectively.
One of the objectives of the proposed approach is to detect the location of both the fault
and DGRESs in the system and then form the sub-island. According to the proposed
approach, the DGRESs can operate as power islands to maintain un-interruptible power
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 51
supply for customers with high priority. Fig. 3.4 shows an outline of the proposed sub-
islanding scheme for a distribution network with multiple DGRESs.
An important part the proposed scheme is how to determine the location of the faults.
Then, after locating the DGRESs, according to the topology of the distribution network
and the function of the protective devices, the network can be divided into different sub-
islands, as shown in Fig. 3.5. A sub-island should be formed such that it has a reasonable
balance of load and DGRESs. As shown in Fig. 3.5, to implement the proposed approach,
different types of protective devices need to be installed in the system. First, circuit
breakers (CBs), installed to disconnect the DGRESs from the network need to be fast, have
sequential close and open capability, and be able to receive and act on remote close and
open commands. To prevent any out-of-synchronism, these CBs should be equipped with
check-synchronization functionality. Secondly, a computer-based relay with calculation
and data saving capabilities needs to be installed in the substation. The main objective
of using this relay is to distinguish the type and location of faults for delivering proper
commands to the protection devices. Such protective devices are available in the market
[101], [102]. In the case of temporary faults, reclosing would be performed by the main
relay itself.
This sub-islanding approach includes an offline step for data generation and a number
of online functions, with the first being to identify the exact fault location in terms of the
fault line number (Step 1). The second step involves using the backtracking algorithm to
determine the location of DGRESs in the network (Step 2). The third step is to define
and implement the sub-island in conjunction with the protection coordination strategy to
isolate the faulty section (Step 3). Finally, reliability indices are calculated for the system
using the FMEA method (Step 4).
Data generation
Identify location of
DGRES by Step 2
backtracking
algorithm
Main protection
coordination
algorithm
Perform sub-island
Reliability analysis
Step 4
Sub-island is Sub-island
not allowed is allowed
according to the utility practices [18]. Each fault location in the network can be charac-
terized by the current contribution from each source in the faulted network. Therefore,
these data can be used to train the RBFNN to find out the mathematical function of the
distribution network. Meanwhile, the short circuit analysis should be done for all types
of faults involving different phases in all step points. Then, the information bank should
be provided containing a proportion of the current contribution from each source to each
type of fault at each step point required to train the RBFNN using the maximum current
of the three phases related at each step point.
The DIgSILENT Power Factory software version 15 is used to simulate various fault
locations for each line. In the situation of a fault in the system, the currents flowing
through the main source and all DGRESs are chosen as input variables for the RBFNN.
The fault distance from each source and the fault line number are considered as outputs
of the RBFNN. However, after any significant change in the load, DGRES or system
configuration, the load flow and the short circuit analysis need to be updated [93].
The additional offline data required for proper operation of the protective devices in
the system are as follows:
• estimated average load for all load points of the network and their priority in terms
of implementing load shedding
In summary, through the offline calculation, 8 RBFNNs (Fig. 3.6) are trained with
the proper input data, which are gathered by system modeling, load flow analysis and
performing short circuit calculation in different locations. In the case of a fault in the
network, through the online procedure, the accurate type and location of a fault will be
determined.
Protective devices (mainly fuses) have traditionally been used to determine faulted sections
of a distribution network [93]. As the short circuit fault is the worth case of the faults in the
system, this dissertation is focused on this type of fault. Coordination between protective
devices may be lost when DGRESs are added to a network. A procedure to determine
the fault location in a multisource transmission line was proposed in [103]; however, the
network was only a small subset of a multisource network. In [104] and [105], the methods
for fault detection in interconnected transmission networks are discussed. However, the
methods address only fault location after the protective device has operated and therefore
may not be improved for an on line fault location strategy in distribution networks.
In a normal operation condition, the sum of the current phasors from the main source
and all the DGRESs would be equal to the total load current in the system. In the
case of a fault in any part of the system, the fault current would exceed the total load
current. This can be used to detect a fault condition, with the relay sensing a fault in the
distribution network. Before determining the exact location of a fault in the system, it is
necessary to determine the fault type. In this chapter, four types of faults are considered
in the training procedure of the RBFNN (Table 3.2). Among all these types of faults,
the 1-Phase to ground fault has the most possibility of occurrence around 70-85% and the
3-Phase fault has the worst effect on the network which are considered and tested by the
proposed scheme. To identify the fault type, the three-phase currents of the main source
are used. These can be determined by normalizing the three-phase output current of the
feeding substation as follows:
Start
IS IS
IDGRESs
Normalization by
Eq.(13)
Generation
No
1 phase
to ground fault
Yes
RBFNN 1
No
2 phase fault
RBFNN 2
Yes
RBFNN 3 2 phase No
to ground fault
RBFNN 4 Yes
RBFNN 5
3 phase fault
RBFNN 6 Yes
RBFNN 7
RBFNN 8
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Faulty line
55
End
ns
X
If = (If )sources (3.15)
s=1
If
Inormal = (3.16)
Imax
where If is the fault current and Imax is the maximum fault current for each type of the
fault. The type of fault current can be obtained by using (3.16) and Table 3.2.
Note that, as the fault resistance affects the accuracy of the short-circuit location, a
method is taken from [106] to eliminate the influence of the fault resistance in determining
the distance to the fault.
In this dissertation, a two-stage RBFNN is used to identify the exact location of the
fault. The procedure to determine the on-line fault location is as follows:
1- Three-phase currents of all sources in the system are used as inputs for the first
RBFNN.
2- The outputs of the first RBFNN, which are the distances of fault from the sources,
are used as inputs for the second RBFNN.
3- The outputs of the second RBFNN represent the exact fault location.
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 57
12 13
Backward track
L12 LP7 Forward track
LP6 DGRES2 LP Load Point
L10 L11
LP5 11
L Line
10 L9
Main 1 2 3 4 5
Supply LP4
L2 L4 L5
L1 Faulty line
L6 L3
LP3
6 L13
7 8 9
L7 L8
LP2
LP1
DGRES1
DGRES3
Figure 3.7: The backtracking search algorithm in a sample distribution network with
DGRESs.
After identifying the faulty section, in order to prevent the DGRESs from shutting down,
it is necessary to find the location of the DGRESs. First, if a DGRES is located in the
faulty section, its protection should trip the associated CB within two seconds of island
formation. Second, make sure that any possible changes after offline calculation can be
covered by the proposed algorithm. In fact, if a new DGRES is added to the network, the
algorithm can handle this situation by finding the proper location of the new unit.
In this chapter the backtracking search algorithm is used to find the location of the
DGRESs [96]. Fig. 3.7 demonstrates the procedure within a sample distribution network.
From Fig. 3.7 it can be seen that the backtracking algorithm begins at the faulty line,
which is Line 3, and searches for the DGRESs in the next buses using Lines 1, 2, 4, 6 and
9 as five independent paths. If the detected bus in each path does not have any DGRES,
the next bus in the path is then checked. In this example, this procedure continues until
the paths leading to the main source, DGRES1, DGRES2 and DGRES3, are detected.
The algorithm uses backtracking at Buses 5, 8 and 10 because each of these buses is the
last bus in the path without DGRESs.
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 58
After completing all the aforementioned procedures the sub-islands must be defined to
allow the DGRESs to supply customers within these zones. In the proposed scheme, the
network is divided into several sub-islands according to the distribution network topology
and protective device location, in terms of the following three categories (Fig. 3.5):
1) Sub − island − S (SIS ): is defined as the sub-island where the load points (LP) can
be restored by the substation as soon as the fault is isolated.
2) sub − island − N (SIN ): includes the sub-islands that have no DGRES, whose loads
remain unsupplied during the fault period and can only be restored by the substation after
repairing the faulty line section.
3) sub−island−DG (SIDG ): includes the sub-islands that have one or more DGRESs.
The LPs located in this sub-island can be restored by the DGRES during the repair period
of a faulty line section. In this situation, the controllers of the DGRESs are assumed to
allow two modes of operation: grid-connected and islanded [107].
After determining the faulty line using the RBFNN and the DGRES locations using the
backtracking algorithm, the main protection coordination algorithm starts by identifying
all the downstream paths leading to DGRESs. The relay then sends a trip signal to
isolate the faulted sub-island by opening the protective devices at the faulty line and the
protective devices attached to the immediate downstream bus leading to the DGRESs.
To enable restoration in the case of a temporary network fault, one CB for each sub-
island should be recognized in advance as a sub-island forming breaker. The relay can
then perform a reclosing action on these CBs, depending on which sub-island is faulted.
Since the current contribution of the sources can be monitored continuously by the relay,
after each reclosing action, it would immediately detect whether or not there was a fault.
If the fault has disappeared, the relay would send closing signals to each CB that needs
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 59
Start
System modeling A
Fault within
DGRES location Yes Shut down DGRES
DGRES zone
Isolate
Number of Run reliability
NO faulty Yes End
DGRES paths >0 assessment
Line?
Yes No
Open all the relay attached
Protection
to bus leading to DGRES
coordination is lost
and relay at line N
to fulfill synchronized closing. This closing can be detected by the relay and the CB of
DGRES would be closed and the system restored to normal operation.
In the case of a permanent fault, the sub-island zone may experience an outage time
equal to the repair time of the faulted section. In this case there is a possibility that more
sub-islands would form after isolating of one sub-island. In the case of an unintentional
island or loss of main (LOM) (CB1 off in Fig. 3.5), depending on the location of the fault,
there would be multiple sub-islands in the network.
The flow chart of the main protection coordination algorithm is depicted in Fig. 3.8.
To show how the main protection coordination algorithm works, a sample distribution
feeder is considered (Fig. 3.5). It is assumed a single fault occurring in Line 2. After
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 60
determining the faulty line and the downstream paths leading to the DGRESs by the
aforementioned algorithms, a signal is sent to the protective devices located in this faulty
line: R2, R3 and the CB for the DGRES. By tripping all these devices, the faulted Line
2 will be isolated by the proposed algorithm, DGRES2 can operate in island mode and
supply the loads. Hence, in the case of LOM, both DGRES1 and DGRES2 can operate in
island mode.
In this study, the FMEA method is adopted to calculate the reliability indices of the power
distribution network [7]. The following are evaluated to assess the impact on the system
reliability: SAIFI and SAIDI. In addition, the following system indices are also evaluated:
ENS, ECOST and interrupted energy assessment rate (IEAR)[7]. The failure rate and
repair time for a sub-island zone can be calculated as follow.
Nc
X
λSI
m = λim (3.17)
i=1
Nc
SI 1 X
rm = SI λim rm
i
(3.18)
λm
i=1
NSI
1 X SIS SIN SIDG
SAIF I = λSI
m (FLP + FLP + FLP ) (3.19)
NLP
m=1
NSI
1 X SIS SIN SIDG
SAIDI = λSI
m (OLP + OLP + OLP ) (3.20)
NLP
m=1
N Nq NLP NCT
SI X
X XX
ECOST = λim × CDFijkm × Ljkm (3.21)
m=1 i=1 j=1 k=1
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 61
The interruption costs for the given type of customers as a function of interruption are
presented in Table 3.3.
The frequency of interruption and associated outage time of the LPi inside of each
category of sub-islands can be calculated as follows:
SIS SIS
OLP = ts NLP (3.22)
SIS SIS
FLP = NLP (3.23)
SIS
where ts is the switching time for isolation of the faulty section, and NLP is the number
of LPs that can be supplied by the substation after isolation of the faulty section.
2) Calculation f or SIN : The LPs that can olny be supplied by the substation after
repairing the faulty section will experience the outage time based on the repairing hours of
SIN SIN
the faulty section. Thus, the LP outage time OLP and LP interruption frequency FLP
associated with the SIN can be calculated as :
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 62
SIN SIN
OLP = tr NLP (3.24)
SIN SIN
FLP = NLP (3.25)
SIN
where NLP is the number of LPs that cannot be supplied by the substation or DGRES.
3) Calculation f or SIDG : The LPs that can be reconnected to the DGRES, may be
supplied based on the restoration strategies and the available capacity of DGRESs. Thus,
SIDG SIDG
the LP outage time OLP and LP interruption frequency FLP associated with the SIDG
can be calculated as :
SIDG SIDG
OLP = ts NLP (3.26)
SIDG SIDG
FLP = NLP (3.27)
SIDG
where NLP is the number of LPs that can be supplied by the DGRESs after isolation
of the faulty section.
Note that, If the DGRES does not have sufficient generation capacity to meet the
load demand over the restoration period in the sub-island, load shedding strategies (such
as curtailment of the LPs with lower customer interruption cost) will be implemented to
alleviate generation-demand mismatch. A simple structure has been considered for the
load shedding which is based on the customer types and CIC for each customer. Based
on this procedure, the customers with lower CIC will be shed first.
is greater or equal to the load demand in the sub-islanded zone and protective devices
operate successfully.
The probability of a sub-island being successful when there is a failure of any component
can be determined from the following equation [16]:
X
SIps = {( (PGDGRES × PGLP )) × (1 − Pf )}{PCB } (3.28)
In case that there is no DGRES in the sub-islanded zone the probability of successful
operation can be calculated as follows:
In this chapter it is assumed that the load points in the sub-island will be supplied
continuously when the sub-island is forming. Considering that the load point LPi is in the
sub-islanded zone with a probability of SIps , the failure rate and associated outage time
of LPi inside the sub-island can be calculated as follows:
SI SI
OLP =(OLP |sub − island sucessf ul) × SIps
(3.30)
SI
+ (OLP |sub − island f ails) × (1 − SIps )
SI SI
FLP =(FLP |sub − island sucessf ul) × SIps
(3.31)
SI
+ (FLP |sub − island f ails) × (1 − SIps )
operation of DGRESs will establish continuity of supply to critical load points whereby
the DGRES should have enough capability to supply the entire sub-island. Otherwise, in
the case of line overload some loads must be shed. Consequently, the proposed approach
helps to enhance the reliability of supply to customers.
Feeder 4 of the 33kV distribution system of the Roy Billinton test system (RBTS) with
three DGRESs is used for the simulation [109]. Fig. 3.9 shows the single line diagram, and
the location of protective devices and DGRESs. Required data, such as component failure
rates, average and maximum load, number and type of customers on feeders etc., are given
in [109]. Three DGRES technology options, namely biomass generator, wind turbines and
solar PV are considered. The probability function for the PV and wind turbines output
power and network loads are extracted from [108].
The data required for training the RBFNNs, which is performed in an offline calcula-
tion, were generated by simulating various types of faults occurring at each 100 meters of
each system line on an Intel(R) Core(TM) i5-2500 3.3 GHz processor with 8 GB of random-
access memory (RAM). The MATLAB software is used to estimate the fault distance from
each source, faulty line section and location of DGRESs, respectively. The aforementioned
protection strategy is also written in MATLAB codes to perform the restoration. 3654
training and testing data for fault types were generated. 80% of the data were used for
training and 20% for performance testing. Table 3.5 shows the results of the trained
RBFNNs in terms of their mean squared errors (MSE) relative to the goal. The MSE
errors show that the training procedure was successful and the algorithm is capable of
finding the fault and DGRES location.
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 65
33kV
CB1
CB
L35
Sub-Island1 Load Point (LP)
LP18
L36 Sub-Island zone
LP19
L37
LP20
L38
LP21
L39
LP22
L40 L41
LP23
L43 L42
LP24
CB2 L44
LP31 L53 CB3
L54 L45
LP25
LP32 L55
L46
L56 L47
LP26
LP33
L48
L57
LP27
LP34 Sub-Island5
L58 L49 CB5
L59 LP36
LP35
CB4
L60
L62 L63 L64
DGRES1
L61
Sub-Island2 L50
DGRES3
LP28
L51
LP29 Sub-Island4
LP30
DGRES2
Sub-Island3
Figure 3.9: Radial distribution test system (RBTS, Bus 6, feeder 4).
Table 3.5: Training Performances of the RBFNN for the Test System
The first reliability analysis (Case 1) assumes the standard mode of operation in which
islanding is not allowed. Consequently, some of the customers experience an outage time
equal to the repair time of the faulty component. The second reliability analysis (Case
2) assumes the proposed sub-islanding scheme. For instance, when a fault occurs in Line
46 (L46), the main protection algorithm determines three paths. Each path starts from
the bus immediately upstream or downstream of the faulty line, with the end of each
path being the DGRES or the main source. After determining all paths that lead to the
DGRESs, the main protective relays are operated to isolate the faulty line (L46). To do
this, CB3, CB4 and CB5 are tripped to isolate this faulty section. As a result, there are
two sub-islands in the system and DGRES2 and DGRES3 can stay online while operating
in the island mode. There is a possibility of making four sub-islands when a fault occurs in
L38. In this situation, DGRES1, DGRES2 and DGRES3 can supply the loads in the sub-
islanded zones. There is one sub-island without a DGRES and customers in this sub-island
would experience the outage time equal to the repair time of L38.
The reliability results for the modified test system are shown in Table 3.6 and Table 3.7.
The effectiveness of the proposed approach is documented in the presented results. 50%
improvement in ECOST and ENS can motivate decision makers to use this approach
when planning measures to improve the reliability of power distribution systems. As can
be seen in Fig. 3.10, the proposed approach is also seen to improve the load point energy
not supplied (LPENS). The load points that are furthest from the main supply receive the
most benefit from the introduction of DGRESs into the system.
Also, to verify the proposed analytical approach, the results of Case 2 are compared
with those obtained by the Monte Calro simulation (MSC) and presented in Table 3.6.
In the MCS, the load demand and generation pattern are the same with the analytical
method [108]. The exponential distribution functions are implemented to simulate the
status of DGRESs and system components [110]. Furthermore, the normal distribution
function with standard deviation of 30 minutes is used to generate the switching and
repair time. By comparing the results in Table 3.6, it can be seen that the reliability
indices derived by the analytical method and MCS are approximately the same with very
small differences. The maximum differences between SAIFI, SAIDI, ENS, and ECOST
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 67
Table 3.6: Reliability Indices of the Modified Test System for Two Cases
Table 3.7: LPENS results for modified test system (Case1: according to IEEE 1547-2003
standard and Case2: according to the proposed approach)
obtained by the analytical method and those obtained by MSC are 0.09%, 0.33%, 0.39%,
and 0.75%, respectively.
The running times of the proposed algorithms are presented in Table 3.8. The results
demonstrate that the proposed approach is fast and meets the requirements of IEEE
Std. 242-2001 [111] and 1547-2003 by detecting the fault and DG position in less than
two seconds. The training and testing time of the data set depend on the component
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 68
LPENS (MWh/year)
2
Case1
1
Case2
0
LP18
LP19
LP20
LP21
LP22
LP23
LP24
LP25
LP26
LP27
LP28
LP29
LP30
LP31
LP32
LP33
LP34
LP35
LP36
LP37
LP38
LP39
LP40
Figure 3.10: LPENS results for the modified test system.
Algorithm time(s)
Offline training data 210
Online algorithms 0.189
of the system and the input data of the RBFNN. Also, the advantages of the protection
coordination scheme in comparison with two of the existing conventional artificial neural
network (ANN) based schemes are shown in Table 3.9.
To assess the effects of the number and location of DGRESs on the reliability indices, two
sensitivity analyses were conducted on the test system.
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 69
90
80
70
ECOST(k$/year) 60
50
40
30
20
10
0
DGRES3 DGRES2 DGRES1 NO DGRES
Depending on the location of DGRES, some load points that are located in the sub-island
can benefit from the presence of DGRESs. The impacts of the location of the DGRESs on
both the load point and system reliability indices were analyzed. To do this, one DGRES
was added to the system in each sub-island. For comparison, simulation was performed
without any DGRES as a base case. Variation in ECOST versus the location of each
DGRES in the system is plotted in Fig. 3.11. The results indicate that adding DGRES3
has the gratest effect on improving ECOST. This is because this part of the test system
consists of five load points with higher load levels. On the other hand, feeder capacity and
customer type are the key parameters for selecting the proper location of DGRESs in the
network.
By increasing the penetration of DGRESs in the system, the benefits of the proposed
method become more apparent. To demonstrate the superiority of the proposed sub-
islanding approach, a sensitivity analysis investigating the impact of increasing the pen-
etration of DGRESs was performed. The variation in ECOST versus the number of
DGRESs in the system is plotted in Fig. 3.12. Importantly, the first DGRES dramatically
enhances the reliability indices of the system when the proposed method is used. In the
Chapter 3. Distribution System Reliability Analysis Considering Distributed
Generation 70
90
80
70
ECOST(k$/year) 60
50
40
30
20
10
0
NO DGRES 1DGRES 2DGRES 3DGRES
test system, when DGRES3 is included as the first DGRES, the ECOST value is improved
by around 24%. This sensitivity analysis also verifies the effectiveness of the proposed
approach in the cases of high penetration of DGRESs.
3.7 Conclusion
A sub-islanding approach based on two algorithms has been presented in this chapter. The
first algorithm implements a network zoning approach to defining the sub-islands, in which
each sub-island is treated as an independent section and is capable of independent island
operation when needed. In addition, the implementation time calculated for the proposed
approach is quite fast and meets the requirements of IEEE standards. The proposed
approach has been successfully tested on feeder 4 of the RBTS bus 6. Numerical results
and a sensitivity analyses have revealed the efficient performance of this sub-islanding
approach. The sensitivity analysis also quantified the benefits associated with changing
the number and location of DGRESs in the system.
Chapter 4
71
List of Symbols
Symbol Definition
Ljtf Annual average load at Load Point j of Feeder f at Year t
δ Annual load increase rate
xqs Binary decision variable
cγ Cost coefficient
Cjqf Customer outage cost continuous decision variable
rb0 qs Customers outage cost value for the outage duration equal to switching time
rb00 qs Customers outage cost value for the outage duration equal to equipment
repair time
ICTqjtf Interruption cost of Load Point j due to temporary outage in Section q of
Feeder f at Year t
Ctemp Interruption cost ($/kW) of temporary faults
T Life expectancy of a sectionalizing switch
m Number of constraints
n Number of coefficients of the cost function
z Number of elements in Z
aιγ Parameter to describe the constraint
bι Parameter to describe the constraint
CCs Sectionalizing switch capital investment cost
ICs Sectionalizing switch installation cost
M Cs Sectionalizing switch annual operation and maintenance cost
Z Set of product term
ECOST System expected interruption cost of customers
72
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 73
λ0qjtf Temporary failure rate of Load Point j due to the failure in Section q of
Feeder f at Year t
TC Total cost of reliability
SC Total investment, installation and operation cost of sectionalizing switches
Nf Total number of feeders
Nq Total number of possible fault locations
NLP Total number of load points
Ns Total number of possible locations to install sectionalizing switch
Nas Total number of available switches
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 74
4.1 Introduction
Introducing DG into distribution networks can affect the problem of optimal switch place-
ment. A method for solving this problem of optimal placement of sectionalizing switches
with a large number of DG units is presented in this chapter. The primary objective is to
minimize the total cost of achieving a certain level of reliability in the presence of DG. The
problem is formulated using mixed-integer linear programming to determine the optimal
number and location of sectionalizing switches. Allowing DG to operate in an island mode
is also considered. The effectiveness of the proposed formulation is evaluated on a modi-
fied reliability test system. The presented results indicate that the proposed formulation
can provide a global optimum solution for the problem of sectionalizing switch placement,
while reliability, capital investment and annual operation and maintenance costs are con-
sidered. Furthermore, DG can play a key role providing a reasonable improvement in the
reliability indices of the system.
The need to operate electricity distribution networks with high reliability and efficiency
while minimizing associated costs remains a challenge for utilities and network planners.
The main challenge is that increasing levels of reliability requires greater capital and
operational expenditure [114]. Optimization is therefore required to find a balance between
reliability improvement and cost.
According to [7], about 80% of customer supply interruptions are associated with fail-
ures in the primary distribution network. The frequency and duration of interruptions
are important reliability indices associated with the distribution networks. A significant
amount of resources is invested on improving these indices to improve the supply avail-
ability and reduce restoration times.
An effective way to improve these indices is to use sectionalizing switches. Both the
number and location of sectionalizing switches need to be optimized [7].
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 75
DG has and will continue to challenge existing protection designs while the placement
and type of protective devices affect the reliability indices. DG can potentially improve
network reliability, as it can reduce interruption duration and restoration time after a given
fault. However, such improvement depends on the capability of DG systems to operate in
islanding mode. An island can be formed when sufficient local generation exists to avoid
load prioritization and/or shedding.
Therefore, the problem of optimum switch placement optimization becomes more com-
plicated when DG units are present in the distribution network. It is quite often difficult to
determine the optimal position of sectionalizing switches for supporting the power supply
of the islanded area in the presence of faults. A number of algorithms have been proposed
to solve the switch optimization problem in distribution networks with DG. In [19], an
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 76
optimization approach based on the ant colony system (ACS) algorithm was developed
to determine the optimal recloser and DG locations by minimizing a composite reliabil-
ity index. Another method for sectionalizing switch placement using an ACS algorithm
was developed in [121]. The authors proposed a fuzzy multi-objective approach to solve
the problem. A genetic algorithm method for simultaneously allocating DG units and
automatic switches in power distribution networks was presented in [20].
Excluding a few exceptions, the heuristic algorithms explore only a limited region
of the search space and have a tendency to be stuck on a locally optimal solution [12].
However, mathematical optimization methods such as linear/nonlinear programming and
MILP, have been demonstrated to reach the global optimum solution in a finite number
of steps [23], [19].
Heuristic techniques, such as those based on priority lists, have not totally succeeded,
as they often lead to low quality solutions. Metaheuristics had very promising outcomes
when they were first explored. The quality of their results was better than those achieved
using well established techniques, and the solutions were obtained very quickly. However,
some drawbacks can be highlighted when metaheuristic techniques are used. If one con-
siders that the ultimate goal is to design a technique that can be accepted and used by
a company, one major drawback of the metaheuristic techniques is their dependence on
parameter tuning. Parameter tuning is time consuming and the complex tuning procedure
requires profound knowledge of the implemented algorithm. Furthermore, accurate tuning
is vital for algorithm performance. A second drawback is related to the lack of information
that metaheuristics provide in terms of the quality of the solution (i.e., how far it is from
the optimal solution). Some proposals have been presented to address these drawbacks;
but this still remains an open area of research.In contrast, mixed-integer nonlinear pro-
gramming (MINLP) guarantees convergence to the optimal solution in a finite number
of steps while providing a flexible and accurate modeling framework. In addition, during
the search of the problem tree, information on the proximity to the optimal solution is
available. Currently, the dramatic increase in efficiency of mixed-integer programming
(MIP) solvers encourages thorough exploitation of their capabilities. Efficient software
for mixed-integer linear and nonlinear programming has been developed, and optimized
commercial solvers with large-scale capabilities are also available.
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 77
The objective of this research study is to propose an MILP-based method for solving
the problem of optimal sectionalizing switch placement in the presence of DG. In this
chapter, based on the islanding operation of DG, which was reported in Chapter 3, a
formulation for calculating the outage time of the load points inside and outside the island
is reported in detail. As the outage time of the customers is very important to evaluate the
cost of reliability, this formulation can provide a precise assessment of reliability indices.
This formulation has not been considered in the switch optimization problem with MILP
formulation yet. The most common customer-oriented reliability indices such as SAIFI,
SAIDI and ENS, were used in [19]-[20] to assess the optimal number and location of
sectionalizing switches in distribution networks in the presence of DG. But these indices
are not sufficient to represent the outage cost of customers [2, 23]. Therefore, in this
chapter, the ECOST index is used to associate reliability with customer outage costs
since it accounts for the effects of the interruption durations, failure rate of equipment,
load variations, customer types and customer damage functions. Furthermore, in this
chapter, the cost of interruption due to the temporary faults is considered in the proposed
MILP formulation, which was not considered in the cited references except in [114]. In this
chapter, the objective is to find the optimal number and location of sectionalizing switches
to minimize the outage cost of customers while sectionalizing switch capital investment,
installation and annual maintenance costs are considered in the proposed formulation.
Several sensitivity analyses to assess the effects of different parameters on the optimization
problem are also reported. Furthermore, the impact of DG on the optimal number and
location of sectionalizing switches is discussed.
4.3.1 Fundamentals
The benefits of sectionalizing switches can be expressed in terms of reducing the duration
and frequency of interruptions. This can be achieved by isolating faulted sections of the
network and restoring power to the healthy sections of the network. However, the network
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 78
topology, the number and type of customers, and extra protective devices can compromise
the expected benefits of sectionalizing switches.
In this study, the problem of identifying the optimal number and location of section-
alizing switches in a distribution network is modeled as a MILP problem. The objective
is to minimize the customer interruption cost while considering the capital investment,
installation and maintenance costs of the sectionalizing switches.
A linear function with linear variables subject to linear constrains can be optimized
by integer programming. A mixed integer program is when some, but not all, variables
are restricted to be integer [2]. The branch-and-bound approach is applied to solve the
following problem, in which some, but not all, variables are constrained to be integer:
n
X
min cγ xγ (4.1)
γ=1
subject to:
n
P
aιγ xγ ≤ bι ι ∈ {1, 2, ..., m}, xγ ∈ {0, 1}
γ=1
where cγ are the cost coefficients, aιγ and bι are the parameters that describe the con-
straints, m represents the number of constraints and n is the number of coefficients of the
cost function.
X
xγ − xn+1 ≤ z − 1
γ∈Z
X
zxn+1 ≤ xγ (4.2)
γ∈Z
xn+1 ∈ {0, 1}
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 79
where Z is the set of product terms and z illustrates the number of elements in Z.
Nf T Nq NLP
X XX X
T C = SC + (ECOSTqjtf + ICTqjtf )(1 + δ)t−1 (4.3)
f =1 t=1 q=1 j=1
SC is defined as the switch cost including capital investment cost (CC), installation
cost (IC) and maintenance cost (MC), and is expressed by:
Ns
X Ns
T X
X
SC = (CCs + ICs )(1 − xqs ) + M Cs,t (1 − xqs ) (4.4)
s=1 t=1 s=1
where, if the switches are installed in the location s, xqs = 0, otherwise, xqs = 1.
Also, taking into consideration the load increase rate during the time horizon under
study, the average load of load points in (4.4)is multiplied by (1 + δ)t−1 [23].
The interruption cost for each load point j owing to the outage in Section q of Feeder
f at Year t consists of ECOSTqjtf and the interruption cost due to the temporary faults
(ICTqjtf ), which are expressed in the following equations:
ECOSTqjtf = λ
bqs Φqjtf (b
rqs ) (4.6)
where
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 80
Normal
State
Repair Switching
Repair Switching
State State
Figure 4.1: State diagrams for components in distribution systems.
Component λ
bqs rbqs
Line λl,qs rl,qs
Switch λsw,qs (1 − xqs ) rsw,qs
Transformer λT r,qs rT r,qs
Φqjtf (b
rqs ) = Ljtf Cjqf (b
rqs ) (4.7)
and λ
bqs is the component failure rate, which is presented in Table 4.2, and Φqjtf (b
rqs ) is the
product of the customer damage function and the average load of Load Point j of feeder
f at year t. Cjqf (b
rqs ) is the outage cost in ($/kW) of Load Point j owing to the outage in
Section q of Feeder f with the outage duration of rbqs . Note that, the cost Cjqf (b
rqs ) can
be obtained from the customer damage function.
the proposed approach to assess the reliability of distribution system is shown in Fig. 4.2.
Also, the following assumptions are considered to derive the objective function.
4.3.3 Constraints
The problem of optimal switch placement needs to meet the targeted reliability level.
One of the constraints for the objective function is the number of available switches for
installation:
Ns
X
(1 − xqs ) ≤ Nas (4.8)
s=1
System Data
Customers
Component Data
Failure Rate
Customer to Customers
Transformer Type
Mapping
Loading of
Each LP
Outage Feeder Topology
Duration & Switch Location
Calculate Customers
Interruption Cost
Switch Cost
Formulate the
Objective Function
where rb0 qs and rb00 qs refer to the switching time of the sectionalizing switches and repair
time of faulted equipment, respectively.
Equations (4.9) and (4.10) restrict the continuous decision variable Cjqf (b
rqs ), based on
the number and location of switches and considering the switching time of sectionalizing
devices and the repair time of faulted equipment.
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 83
The modified distribution test system used in this chapter is the radial distribution system
connected to Bus 4 of the reliability busbar test system (RBTS), consisting of 38 load
points and 51 possible switch locations [109]. The modified test system without normally
open switches is depicted in Fig. 4.3. The required reliability data, such as customer data,
component failure data, and average and maximum load at each load point are presented
in [109]. In addition, all switches are assumed to be automatic and have a switching time
of 10 minutes [23]. The capital investment and installation cost of a sectionalizing switch
is assumed to be $4700 [117], [72] and [23]. The maintenance cost of a switch is assumed
to be 2% of the capital investment cost [117], [72] and [23]. The life expectancy of the
switching device is assumed to be 15 years [23]. Also, the load growth rate of the system
is assumed to be 3% [23]. The CDF data are elicited from Table 3.3.
The proposed formulation has been implemented in the GAMS environment and is
solved by using the MILP solver CPLEX 11.0 on an Intel(R) Core(TM) i5-2500 CPU@3.30GHz
with 8 GBs of RAM. As mentioned before, the normally open switches and alternate sup-
ply are neglected in this test system. Table 4.3 demonstrates the results of using the
proposed formulation for the modified distribution test system connected to Bus 4 of the
RBTS. In this case, the TC value over the 15-year time horizon has been obtained by the
MILP proposed formulation to be $1 938 390 while 12 sectionalizing switches are needed.
The optimal location of the switches is also illustrated in Table 4.3. The TC value is high
for this case since the load points located downstream of a fault cannot be supplied owing
to absence of an alternative supply. However, in this case the program execution time was
around 2.5 seconds, which is a very short time for this case.
As mentioned in the previous sections, the presence of DG can reduce interruption duration
and restoration time of the customers. More DGs can provide more opportunities to
improve the reliability indices of the load points in an isolated area and the entire network.
However, the proposed formulation can cope with even high penetration of DGs into the
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 84
LP14 LP17
LP11 LP12 LP13 LP15 LP16
Sectionalizing
switch 25 30
CB 20 22 24 27 29
F3 19 21 23 26 28
LP8 LP9 LP10
14 16 18
F2 13 15 17
1 3 5 7 10
F1
2 4 6 9 12
Substation at Bus 4
8 11
57 59 62 64 67
61 66
F6 50 52 54
51 53 55
45 47 49
Figure 4.3: Modified distribution test system (with 38 load points (LP) and 51 section-
alizing switches).
Table 4.3: Optimal Number and Location of Switches Using the Proposed MILP
TC (k$) 1938.39
ECOST+ICT (k$) 1815.93
SC (k$) 122.45
Proposed Number of Sectionalizing switches 12
5B,10B,15B,17B,26B,36B
Proposed Locations of Sectionalizing switches*
39B,46B,48B,52B,54B,63B
* Every line section has two candidate locations for sectionalizing switch installation
which are shown as B for the beginning of a line section and E for the end of a line
section.
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 85
Substation
Substation
1a) 1b)
1 S1 3 S2 5 S3 7 1 S1 3 S2 5 S3 7
2 4 6 8 2 4 6 8
Substation
1 S1 3 S2 5 S3 7 1 S1 3 S2 5 S3 7
DG DG
2 4 6 8 2 4 6 8
DG DG
Substation
3a)
Substation
3b)
1 S1 3 S2 5 S3 7 1 S1 3 S2 5 S3 7
2 4 6 8 2 4 6 8
DG DG
Substation
Substation
4a) 4b)
1 S1 3 S2 5 S3 7 1 S1 3 S2 5 S3 7
2 4 6 8 2 4 6 8
DG DG
Substation
Substation
5a) 5b)
1 S1 3 S4 S2 5 S3 7 1 S1 3 S4 S2 5 S3 7
2 4 6 8 2 4 6 8
DG DG
Substation
Substation
6a) 6b)
1 S1 3 S2 5 S3 7 1 S1 3 S2 5 S3 7
DG DG
2 4 6 8 2 4 6 8
network and this issue cannot affect the solution procedure. To provide more details
about how DG can improve the reliability of the system, the following sample test system
is considered as an example .
Fig. 4.4(1a) shows the test system when a fault happened in Line Section 5 and
Fig. 4.4(1b) shows the test system after fault isolation and power restoration procedure.
According to the FMEA method [7], to clear this fault first the CB should be opened
and to separate the fault, S2 and S3 need to open. As we do not have any alternative
supply or DG, the power can just be restored by the main substation to the LP1, LP2.
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 86
The outage time of these two load points is equal to the switching time of sectionalizing
switches which is considered 30 minutes in this example. For the LP3 and LP4, which are
interrupted, the outage time is equal to the repair time of the Line Section 5.
In the case of adding one DG unit into this feeder (Fig. 4.4 (2a) ,(2b)), when a fault
happens in Line Section 5, after fault isolation and separation procedures power can be
restored to LP1 and LP2 by closing the main CB and through the main substation. DG
can provide the power to LP4. In this case, only LP3 is interrupted and the outage time
for this load point is equal to the repair time of the line. The outage time for the supplied
customers is equal to the switching time (30 min). The results for this case are compared
with the base case (system without DG) in Table 4.4. As it can be seen, adding DG
can decrease customer outage time and as a result the ECOST decreased. Note that, the
number and location of switches play a key role to benefit of DG introduction into the
system. To investigate further the impact of DG location, the DG is connected at the
node of LP2 (Fig. 4.4 (3a), (3b)). When a fault happens in the Line Section 5, S2 and
S3 should be open to isolate the fault, and consequently LP3 and LP4 remain interrupted
as the DG cannot participate to supply these loads. The outage time for these two load
points is equal to the repair time of the faulty line. Thus, the location of the DG units is
very important to assess the reliability and allows to provide more benefits by adding DG
at the end of the feeder to more load points in the feeder.
To explain the importance of switch location in case of adding DG into the system
either at the end of feeder or in an arbitrary location, the occurrence of a faults in Line
Section 3 is considered (Fig. 4.4 (4a), (4b)). According to the topology of the system
(DG location, switch number and locations), after fault isolation and power restoration
procedure, LP2, LP3 and LP4 will be interrupted. If just one more switch is added into
this system (S4)(Fig. 4.4 (5a), (5b)), after fault isolation and power restoration procedure,
the faulty line can be isolated by opening S1 and S4, and LP1 can be supplied by the main
substation while the power to LP2, LP3 and LP4 can be provided by the DG. Therefore,
the reliability indices for the system in this situation will improve.
In the case of adding two DG units into this feeder (Fig. 4.4 (6a), (6b)), more im-
provement can be observed for the reliability indices depending on in which section the
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 87
DG is added. In this example, as the second DG unit is inserted to the node of Load
Point 2, the outage time of this load point would improve and consequently the cost of
reliability can decrease. Results for this case are also presented in Table 4.4. Note that,
all the achievable benefits of adding DG units into the system depend on the location and
number of switches. For instance, when there is no sectionalizing switch in the system,
even if there is a DG unit added into the system, the outage time of the customers would
not decrease.
X
PIP = ( (PGDG × PGLP )) × (1 − Pf ) (4.11)
where PGDG represents the cumulative probability of the DG units generating power equal
to or greater than a certain level. In this chapter, it is also assumed that the load points
in the island will be supplied continuously when the island is forming. Considering that
the load point LPj can be in the islanded area with a probability of PIP , the failure rate
and associated outage time of LPj inside the island can be represented as follows:
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 88
where λqsj and rqsj are the failure rate and outage time of LPj in the islanded area,
respectively. λqsj and rqsj are the failure rate and outage time of LPj without DG units,
respectively. In addition, λqsj,DG and rqsj,DG are the failure rate and outage time of the
load point LPj when DG units are present, respectively. By substituting (4.12) and (4.13)
into (4.6) and using (4.3) as the objective function, the proposed MILP can be implemented
to minimize the TC.
Given that the DG location is assumed to be at the end of each feeder for the load
point LPj , the fault can be cleared if a switch is installed and operated between the Line
k and the faulty one. Under such an assumption, the outage time of a load point is equal
to the switching time, and the load point may be interrupted for a shorter time. If there
is no switch between the line segment k and the faulty line, then the load point outage
time would be equal to the repair time of the faulty line segment. The calculation for the
outage time of the Load Point j of Feeder f can be derived from the following equation:
j
X j−1
Y j
X j−1
Y
0
rqsj,DG = λl,qs lk rl,qs xqs + λl,qs lk rl,qs (1 − xqs )
k=1 s=k k=1 s=k
NLP f k−1
X Y
+λl,qs lk rl,qs + λl,qs lk rl,qs xqs
k=j+1 s=j
NLP f k−1
X Y
0
+ λl,qs lk rl,qs (1 − xqs ) (4.14)
k=j+1 s=j
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 89
Substation
DG
N LPf
1
i
LP LP LP LP
Sectionalizing CB
switch
Figure 4.5: Typical distribution feeder with arbitrary location for DG.
where lk is the length of the Line Segment k and NLP f is the total number of load points
0
in Feeder f . Also, rl,qs and rl,qs represent the mean time to repair the line fault and the
switching time to isolate the faulty line, respectively.
For load points not within the islanded area formed by the DG, the outage time of the
load points can be calculated as follows:
j NLP f k−1
X X Y
rqsj = λl,qs lk rl,qs + λl,qs lk rl,qs xqs
k=1 k=j+1 s=j
NLP f k−1
X Y
0
+ λl,qs lk rl,qs (1 − xqs ) (4.15)
k=j+1 s=j
i−1
X i−1
Y i−1
X i−1
Y
0
rqsj,DG = λl,qs lk rl,qs xqs + λl,qs lk rl,qs (1 − xqs )
k=1 s=k k=1 s=k
j NLP f k−1
X X Y
+ λl,qs lk rl,qs + λl,qs lk rl,qs xqs
k=i k=j+1 s=j
NLP f k−1
X Y
0
+ λl,qs lk rl,qs (1 − xqs ) (4.16)
k=j+1 s=j
Note that, in the case of adding DG to arbitrary locations along the feeder, when a
fault occurs in a line segment before the DG unit and if a switch is installed between the
faulty line and the DG unit, the fault can be isolated. The DG unit would be able to
supply the load points within the performed island continuously. The outage time of the
load points is equal to the switching time. Or else, the load point outage time would be
equal to the repair time of the faulty line segment. In the case of a fault in a line segment
between the DG unit and load point LPj , even if a switch is installed, the outage time
of the load points is equal to the repair time of the faulty line. When a fault occurs in a
line segment after the load point LPj , the fault can be cleared if a switch is installed and
operated between Line k and the faulty one. Under such situation, the outage time of the
load points is equal to the switching time. Otherwise, the outage time of the load point is
equal to the repair time of the faulty line segment.
To assess the effects of DG units on the optimal number and location of switches, sev-
eral sensitivity analyses are performed on the modified test system. Two DG technology
options, namely wind turbines (consisting two wind turbines rated at 1 MW each) and a
diesel generator (with a rating of 2 MW) are considered. The probability function for the
wind turbine’s output power and network loads are extracted from [122].
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 91
Table 4.5: Impact of DG Location on the Optimal Number and Location of Sectional-
izing Switches
4.6.1 DG Location
The impact of DG unit location on the optimal number and location of sectionalizing
switches is investigated. A DG unit with the ability to supply the loads in the feeders is
introduced into the modified test system. Table 4.5 presents the effects of DG location on
the optimal number and location of switches. The results indicate that adding a DG unit
on the feeders supplying customers with higher CDFs and heavier loads is more effective
for reducing the TC of the network. For instance, adding DG at the end of Feeders
4 and 5 could decrease the TC by about 14% and 2.3%, respectively. This is due to
the fact that the Feeder 4 includes two commercial customers with relatively high CDFs
compared to Feeder 5 and has the highest load level of the entire studied network. It can
be evidenced that adding DG can reduce the objective function even if a higher number of
installed sectionalizing switches is proposed. In fact, a higher number of installed switches
provides a greater reduction of the outage time of customers, and consequently improves
the objective function and the continuity of supply.
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 92
The effect of adding more DG units into the network is evaluated. To do this, by the results
of previous sensitive analysis, the feeders are selected based on the contribution of adding
DG unit in reduction of TC. On the other hand, first feeder 4 is selected as this feeder has
the lowest TC in case of introducing DG. Feeder 1 is the second nominated feeder to add
DG unit because of lower TC compare to the other feeders. The number of DG units is
increased based on this procedure from one to seven to illustrate the effect of increasing the
number of DG units on the optimal switch placement problem in terms of TC. Table 4.6
shows changes in the proposed number and location of installed sectionalizing switches and
changes in TC as the number of available DG units is increased. Furthermore, the proposed
locations for the switches in the case of introducing one, three and seven DG units in the
test network are presented in Figs. 4.6 to 4.8, respectively. As expected, the introduction
of the first DG unit has the highest effect on reducing the TC, with subsequent increases in
the number of DG units having less impact on the TC. For instance, the first DG unit on
Feeder 4 decreases the value of TC by $334 840. Furthermore, the number of sectionalizing
switches increases as the installed number of DG units increases and reaches a maximum
number when seven DG units are connected. In the case of seven DG units, even if the
solution derived from the proposed MILP algorithm is such that seven more switches are
installed compared to the base case (that is, a network without any DG units), adding
more switches enables an increase to the supplied load and a decrease to the objective
function (TC) related to the solution. This sensitivity analysis also verifies the optimality
of the solution achieved in the base case.
To further establish the effectiveness of the proposed formulation, the proposed location of
the first seven switches is investigated. By comparing with the solution derived from the
MILP formulation, it is shown that the customer types and load density at each load point
of the feeder are crucial parameters in distinguishing the optimal number and location of
the switches in the network. Two different scenarios of DG units’ number are considered
in this section.
Devices
Table 4.6: Impact of DG units’ Number on the Optimal Number and Location of Sectionalizing Switches
LP14 LP17
LP11 LP12 LP13 LP15 LP16
Sectionalizing
switch 25 30
CB 20 22 24 27 29
F3 19 21 23 26 28
LP8 LP9 LP10
14 16 18
F2 13 15 17
1 3 5 7 10
F1
2 4 6 9 12
Substation at Bus 4
8 11
57 59 62 64 67
61 66
F6 50 52 54
51 53 55
45 47 49
DG
LP26 LP27 LP28
31 33 36 39 41
F4
32 35 38 40 43
34 37 42
Figure 4.6: Optimal location of sectionalizing switches in the case of one available DG
unit.
In the first scenario, the sequence of switch placement on a system without a DG unit is
evaluated. The results are summarized in Table 4.7. As shown, the location 39B of Feeder
4 is proposed for placement of the first switch. According to [109], the Feeder 4 has the
highest load level and contains two commercial customers with comparatively high CDFs.
Since the CDFs for commercial customers are substantially greater than those of non-
commercial customers, the proposed locations of the first four switches are on feeders with
higher loads and adjacent to commercial loads. The rest of the switches are proposed to be
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 95
LP14 LP17
LP11 LP12 LP13 LP15 LP16
Sectionalizing
switch 25 30
CB 20 22 24 27 29
F3 19 21 23 26 28
LP8 LP9 LP10
DG
14 16 18
F2 13 15 17 DG
1 3 5 7 10
F1
2 4 6 9 12
Substation at Bus 4
8 11
57 59 62 64 67
61 66
F6 50 52 54
51 53 55
45 47 49
DG
LP26 LP27 LP28
31 33 36 39 41
F4
32 35 38 40 43
34 37 42
Figure 4.7: Optimal location of sectionalizing switches in the case of three available DG
units.
placed on feeders with greater loads and adjacent to small users sector. For example, the
fifth switch is placed on Feeder 2, which has a greater load level and CDF than residential
customers.
In the second scenario, one DG unit is added to the network at the end of Feeder
4. As shown in Table 4.7, the first two switches are placed on Feeder 4 to increase the
supplied load and decrease the objective function. The first switch is placed adjacent to
the commercial loads with higher CDF values and the second switch is located in the
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 96
LP14 LP17
LP11 LP12 LP13 LP15 LP16
Sectionalizing
switch 25 30
CB 20 22 24 27 29
F3 19 21 23 26 28
LP8 LP9 LP10
DG
14 16 18
F2 13 15 17
DG
1 3 5 7 10
DG
F1
2 4 6 9 12
Substation at Bus 4
8 11
57 59 62 64 67
61 66
51 53 55
DG
LP26 LP27 LP28
31 33 36 39 41
F4
32 35 38 40 43
34 37 42
Figure 4.8: Optimal location of sectionalizing switches in the case of seven available
DG units.
middle of the feeder to supply more customers in the network. The suggested locations for
the remaining switches verify the performance of the proposed formulation as the sequence
of switch placement meets the reliability constraint requirements.
To demonstrate the advantages of the proposed MILP method, the approach in this chapter
is compared with other ones presented in the technical literature [20],[121],[19],[117],[72],[23],
[114] and [2], and the results are summarized in Table 4.8.
As mentioned before, the cost of reliability is related to the customers outage time.
Since both the permanent and temporary faults can cause outage time, considering the
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 98
2.5
TC
ECOST+ICT
SC
2.15
2.1
1.5
Cost (Million $)
2.05
1.95
1
1.9
1.85
1.8
0.5 0 5 10 12 15 20 25 30 35 40 45 50
0
0 5 10 15 20 25 30 35 40 45 50
Number of sectionalizing switches
Figure 4.9: TC, ECOST+ICT and SC vs. number of available switches (without DG
units).
2.5
TC
ECOST+ICT
SC
2
2.05
1.5
Cost (Million $)
1.95
1.9
1
1.85
1.8
0.5 1.75
0 5 10 12 15 20 25 30 35 40 45 50
0
0 5 10 15 20 25 30 35 40 45 50
Number of sectionalizing switches
Figure 4.10: TC, ECOST+ICT and SC vs. number of available switches (DG unit at
the end of Feeder 2).
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 99
TC
2 ECOST+ICT
SC
1.5
1.9
Cost (Million $)
1.85
1.8
1.75
1
1.7
1.65
1.6
0.5 1.55
1.5
0 5 10 1415 20 25 30 35 40 45 50
0
0 5 10 15 20 25 30 35 40 45 50
Number of sectionalizing switches
Figure 4.11: TC, ECOST+ICT and SC versus the number of available switches (DG
unit at the end of Feeder 3)
2
TC
ECOST+ICT
1.8
SC
1.6
1.4
1.2
Cost (Million $)
0.8
0.6
0.4
0.2
0
0 5 10 15 20 25 30 35 40 45 50
Number of sectionalizing switches
Figure 4.12: TC, ECOST+ICT and SC versus the number of available switches (system
with two DG units at the end of Feeders 1,4)
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 100
2
TC
1.8 ECOST+ICT
SC
1.6
1.4
1.2
Cost (Million $)
0.8
0.6
0.4
0.2
0
0 5 10 15 19 20 25 30 35 40 45 50
Number of sectionalizing switches
Figure 4.13: TC, ECOST+ICT and SC versus the number of available switches (system
with four DG units at the end of Feeders 1,3,4,7)
2
TC
ECOST+ICT
1.8
SC
1.6
1.4
1.2
Cost (Million $)
0.8
0.6
0.4
0.2
0
0 5 10 15 1920 25 30 35 40 45 50
Number of sectionalizing switches
Figure 4.14: TC, ECOST+ICT and SC vs. number of available switches (system with
seven DG units, one on each feeder).
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 101
temporary faults in the proposed MILP formulation provides more realistic assessment
of the customer outage cost, which was neglected in the above mentioned references.
However, the customer outage cost can be expressed by the ECOST index in the best
way compared to the other reliability indices such as SAIFI, SAIDI, and ENS. In this
thesis, the detailed design of the sectionalizing switches is determined by minimizing the
total cost of reliability that comprises sectionalizing switch capital investment, installation
and annual operation and maintenance costs, and customer outage cost. Furthermore, to
benefit further from adding DG into the system, a probabilistic island approach is used
in this chapter, in which the island operation of DG is neglected in some of the above
mentioned references. Note that, the proposed MILP formulation can cope with adding a
large number of DG units into the network since the way that DG can affect the customer
outage time is considered in the formulation while the references [19]-[20] just considered
adding a limited number of DG units.
Case (2-8): System with one DG unit at the end of Feeders 1-8.
Case 10: System with three DG units at the end of Feeders 1,4,3.
Case 11: System with four DG units at the end of Feeders 1,4,3,7.
Case 12: System with five DG units at the end of Feeders 1,4,3,7,6.
Case 13: System with six DG units at the end of Feeders 1,4,3,7,6,2.
Case 14: System with seven DG units at the end of all feeders.
Case 15: System with no DG while the number of switches is increased from 1 to 51.
Case 16: System with a DG unit at the end of Feeder 2 while the number of switches
is increased from 1 to 51.
Case 17: System with a DG unit at the end of all feeders while the number of switches
is increased from 1 to 51.
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
Devices 102
4.7 Conclusion
The effects of DG on the optimal number and location of switches in distribution net-
works are presented in this chapter. A mathematical optimization methodology has been
implemented to solve the optimization of the sectionalizing switch placement problem in
power distribution networks with a large number of DG units. To improve the continuity
of service, the main goal is defined as minimizing the total cost of reliability while in-
stalling the minimum number of switches. The problem has been formulated as a MILP
and solved with one of the commercial solvers in a computationally efficient manner. The
effectiveness of the proposed formulation has been validated by means of case studies on
the distribution system connected to Bus 4 of the RBTS. Numerical results and sensitivity
analyses have confirmed the efficient performance of this formulation. The benefits asso-
ciated with the increasing number of DG units are determined and presented as results
from the sensitivity analysis. Furthermore, the presented results indicate that the total
cost of reliability and the reliability indices of distribution networks can be affected by the
number of sectionalizing switches and their locations.
Devices
Table 4.8: Comparison of the proposed approach with other ones reported in the technical literatures for radial networks
References [20] [121] [19] [117] [72] [23] [114] [2] Proposed approach
MILP 5 5 5 5 5 3 5 3 3
Method
Heuristic 3 3 3 3 3 5 3 5 5
Permanent 3 3 3 3 3 3 3 3 3
Fault Type
Temporary 5 5 5 5 5 5 5 3 3
ECOST 5 5 5 3 3 3 3 3 3
Objective Function
Switch Cost 5 3 5 3 3 3 3 3 3
Load Growth 3 5 5 3 3 3 5 5 3
DG 3 3 3 5 5 5 5 5 3
Islanding Approach 3 5 3 5 5 5 5 5 3
Number of DG Up to four Up to four Up to six 5 5 5 5 5 No limitation
Chapter 4. Distribution System Planning Considering Optimal Placement of Switching
103
Chapter 5
104
List of Symbols
Symbol Definition
AT M H Annual total man hours
γ Annual load increase rate
CCf Fuse capital investment cost
ICf Fuse installation cost
M Cf Fuse annual operation and maintenance cost
M HC Man hours cost
NR Number of replaced fuses
RCf Replacement cost of a fuse
CCs Sectionalizing switch capital investment cost
ICs Sectionalizing switch installation cost
M Cs,t Sectionalizing switch annual operation and maintenance cost
T Sectionalizing switch and fuse life period
NCT Total number of customer types
NLP f Total number of load points in Feeder f
NLP Total number of load points
Naf Total number of available fuses for installation
Nas Total number of available switches for installation
Nf r Total number of feeders
Nf Total number of installed fuses
Nq Total number of contingencies
Ns Total number of installed sectionalizing switches
T M Cf Total maintenance cost of a fuse
105
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 106
5.1 Introduction
It has been estimated that about 80% of the customer supply interruptions are as-
sociated with the failures in primary distribution networks [5]. Thus, considerable effort
has been devoted to reduce the effects of failures by utilizing protective devices such as
circuit breakers, fuses and sectionalizing switches. The proper number, location and type
of such protective devices play an important role in the reliability of distribution networks
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 107
by minimizing the impact of interruptions. In addition, the optimum number and loca-
tion of protective devices are important to minimize capital investment cost and maximize
customer benefits [2].
The optimal placement of switches and protective devices (fuses and CBs) in distri-
bution networks can reduce the annual failure rate and outage duration of customers and
consequently improves the reliability indices of the networks. Among all, the most common
customer-oriented reliability indices, such as SAIDI and SAIFI, were used in [125]-[126] to
assess the effects of allocation of switches and protective devices in distribution networks.
These indices are not sufficient to represent the cost of reliability. However, the ECOST
index can be used to associate reliability with customer costs in this chapter [7].
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 108
Therefore, this chapter proposes a mixed-integer nonlinear formulation to solve the op-
timal allocation of sectionalizing switches and protective devices problem, simultaneously.
According to both technical and economic issues, the proposed method allows minimizing
the cost of reliability while limiting the number of installed switches and protective devices.
Moreover, the customer interruption cost and the costs associated with switches and pro-
tective devices, such as capital investment, installation, annual operation and maintenance
costs are also considered in the proposed formulation.
The innovative contribution of this chapter comparing to the relevant works can be
summarized as follows:
• The use of MINLP formulation for switch and protective device placement,
• Value based formulation considering the cost of customer outage and protective
devices,
The concepts related to distribution networks reliability, protection and operation, required
to define the problem and the proposed formulation, are presented in this section.
Sub1 Sub2
CB S1 S2 S3 S4
L1 L2 L3
T1
: Circuit Breaker
T2 : Tie Switch
LP1
LP3 : Sectionalizing Switch
LP2
: Fuse
LP3n
Figure 5.1: Possible locations for sectionalizing switches and protective devices in radial
distribution network.
sustained interruption. The failures that occur in the distribution networks always lead to
customer interruptions and can be considered as permanent or temporary. The permanent
failures will happen by the most serious events and always cause sustained interruptions,
which require to dispatch a repair crew to handle them. On the contrary, the temporary
failure will clear themselves away after a short period of time. It should be noted that
a temporary failure can cause either a momentary or a sustained interruption. In this
chapter, a failure is considered either as the occurrence of a permanent failure on a feeder’s
line or transformer or an occurrence of a sustained interruption due to the failure of any
component of the distribution network.
Distribution networks can be divided as either radial, spot or secondary. Radial networks
are the most prevalent and the focus of this thesis. The radial distribution network is
basically served by the substation and consists of one or more distribution feeders. Each
feeder is split into one or more line sections and laterals with sectionalizing switches and
protective devices. Since this is the most frequently form of network configuration, each
feeder can operate radially for effective coordination of protective devices, as presented
in Fig. 5.1. In addition, the power supply can be restored from an alternative supply or
neighbour feeders by tie switches (TS).
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 110
In general, to mitigate the effects of failures to a minimum acceptable level, utilities install
protective devices along the distribution network feeders. The protective devices consid-
ered in this chapter consist of a CB with overcurrent and automatic reclosing relays at
the substation, and sectionalizing switches and fuses along the feeders and on the laterals.
Automatic sectionalizing switches can isolate a faulted section after the operation of an
upstream CB. Fuses are low cost protective devices that have interruption capability but
do not have automatic reclosing capability. In this chapter, fuses are installed on laterals
and a constraint is defined to do not put the fuses on the main feeders.
A variety of indices are defined in [7] to evaluate the reliability of distribution networks.
The basic indices normally provide service reliability data from an individual customer
viewpoint and are called single load point indices. These include the load point average
failure rate (λ), average outage duration (r), and average annual outage duration (U ).
The most frequently used customer oriented indices can be calculated using these three
basic indices and are defined as the SAIFI, SAIDI, customer average interruption duration
index (CAIDI), average service availability index (ASAI), and expected energy not supplied
(EENS). These indices are not sufficient to represent the cost of reliability. The ECOST
index can be used to associate reliability with customer costs since it accounts for the
effects of the interruption durations, failure rate of equipment, load variations, customer’s
types and customer damage functions [7]. The analytical based method, which is employed
to calculate the ECOST in distribution networks for a typical feeder, is well described in
[117]:
Nq NLP NCT
X XX
ECOST = λi × Cijk (toi ) × Ljk (5.1)
i=1 j=1 k=1
where λi is considered as the average failure rate of the distribution network equipment
due to contingency i. Cijk (toi ) refers to the outage cost ($/KW) of customer type k at
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 111
load point j due to contingency i with the outage duration of toi which can be obtained
from the CDF. Ljk is the average load of kth type customer located at the jth load point.
1- For the load point (LPj ) connected to the network, obtain the indices of λi , toi and
Ui for each outage event i contributing to its outage.
2- Evaluate the cost of interruption Cijk (toi ) using the corresponding CDF and outage
duration toi .
4- Repeat Steps 2 and 3 for each outage event contributing to load point j.
5.4.1 Assumptions
In this chapter, the following assumptions are considered to derive the proposed formula-
tion.
All outages are classified as first-order failures [2, 7]. The concept and how to calcu-
late the higher order outages by considering the overlapping events are presented in [7].
Note that, the failure rate and outage time of the component in the system need to be
recalculated based on this type of events, which is beyond the scope of this thesis.
1 if a sectionalizing switch is installed on
Xsf r = location s of feeder f r,
0 otherwise.
1 if a fuse is installed on location f of
Yf f r = feeder f r,
0 otherwise.
T C = SC + F C + ECOST (5.2)
where SC is defined as the sectionalizing switch cost including capital investment cost
(CC), installation cost (IC) and maintenance cost (MC) which is expressed by:
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 113
Nf r Ns
X X
SC = (CCs + ICs ) × Xsf r
f r=1 s=1
(5.3)
Nf r Ns
T X
X X
+ M Cs,t × Xsf r
t=1 f r=1 s=1
Nf r Nf
X X
FC = (CCf + ICf ) × Yf f r
f r=1 f =1
(5.4)
Nf r Nf
T X
X X
+ T M Cf,t × Yf f r
t=1 f r=1 f =1
ECOST =
Nf r Nq NLP NCT (5.5)
X XX X
λif r × Cijkf r (toi ) × Ljkf r (1 + γ)t−1
f r=1 i=1 j=1 k=1
When a permanent fault occurs in a section or component of the feeder in Fig. 5.1
(e.g., L2), the CB in the substation operates and de-energizes all downstream customers.
Upstream restoration is performed by opening the adjacent upstream sectionalizing switch
(first switch upstream from the fault, eg., S2). This provides service restoration for the
upstream customers. The outage duration for these customers is equal to the switching
time (ts ). In downstream restoration, the adjacent downstream sectionalizing switch is
opened and the normally open tie switch (TS) is closed, allowing downstream customers to
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 114
be fed by adjacent feeder [8]. The outage duration for these customers would be ts + tTs S ,
where tTs S is the switch time of TS. The customers located in the faulty section will
experience an interruption with a duration equal to the repair time of the faulty Section
tr . Therefore, the calculation of outage duration of the load point j is as:
j
X j−1
Y j
X j−1
Y
toj = λi tri Xsf r Yf f r + λi ts (1 − Xsf r Yf f r )
i=1 s,f =i i=1 s,f =i
NLP f i−1
X Y
+ λj trj + λi tri Xsf r Yf f r (5.6)
i=j+1 s,f =j
NLP f i−1
X Y
+ λi (ts + tTs S )(1 − Xsf r Yf f r )
i=j+1 s,f =j
In the case of no TS in the network, the calculation of outage duration of the load
point j is:
j
X j−1
Y j
X j−1
Y
toj = λi tri Xsf r Yf f r + λi ts (1 − Xsf r Yf f r )
i=1 s,f =i i=1 s,f =i
NLP f
(5.7)
X
+ λi tri
i=j+1
5.4.3 Constraints
This section presents technical and economic constraints that are incorporated to the
proposed MINLP model. The technical constraints are related to the network topology and
utilities practice. For instance, if the installation of a CB is mandatory in the substation
by utility practice, the constraint X1 = 1 must hold (the subscript 1 shows the first
line section of a feeder, defined as the substation). In fact, if it is necessary to install a
particular device in a given section, the corresponding variable should be set to one. The
constraints given by:
ts
Cijkf r (toi ) ≥ CDFijkf r × (1 − Yf f r ) (5.8)
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 115
sj
X
tr
Cijkf r (toi ) ≥ [CDFijkf r × (1 − Xsf r )] × (1 − Yf f r ) (5.9)
s=si
are supposed to restrict the continuous decision variable Cijkf r based on the number and
location of switches and protective devices considering the switching time of sectionalizing
devices, actuation time of fuses, and repair time of faulted equipment.
In fact, Cijkf r (toi ) is a set of positive continuous variables and depends on the location
and number of installed switches and protective devices. To show how the constraints
(5.8) and (5.9) can restrict the customer outage cost based on the existence or inexistence
of the switches and protective devices between the contingency (fault) location and load
points, a simplified feeder shown in Fig. 5.1 consisting of two line segments (L1, L2)
and 2 load points (LP1, LP2) is considered. There are three and two possible locations
to install sectionalizing switches (S1-S3) and fuses, respectively. Also, because this test
system has two line segments and two transformers (T1, T2), there are four possible fault
locations (fault in L1, L2, T1 and T2). Based on the (5.8) and (5.9), for instance, if a fault
occurs in transformer (T1), which is connected to the load point 1 (LP1), the following
ts
set of constraints can be defined to restrict the continuous variables. Note that, CDFijkf r
tr
and CDFijkf r are fixed values and present the CDF values corresponding to the outage
duration equal to ts and tr , respectively. The LP1 is assumed to be residential, thus k and
f r stand for 1.
ts
C2111 (to2 ) ≥ CDF2111 (5.10)
tr
C2111 (to2 ) ≥ CDF2111 (5.11)
ts
C2211 (to2 ) ≥ CDF2211 × (1 − Y11 ) (5.12)
tr
C2211 (to2 ) ≥ CDF2211 × [1 − X21 − X31 ] × (1 − Y11 ) (5.13)
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 116
The economic constraints need to be aggregated to limit the utilities costs to main-
tain (or improve) reliability in the case of budget inadequacy. Hence, the sectionalizing
switches and protective devices are comparatively expensive, thus by increasing the num-
ber of installed switches in distribution networks the SC, FC, and CBC costs will increase
dramatically. Nevertheless, it is necessary to optimize the number of these devices ac-
cording to a budget limitation while maintaining (or improving) reliability indices in a
reasonable level. The economic constraints to restrict the number of available devices for
installation in the case of budget inadequacy are as following:
Nf r Ns
X X
Xsf r ≤ Nas (5.14)
f r=1 s=1
Nf r Nf
X X
Yf f r ≤ Naf (5.15)
f r=1 f =1
To ensure that no more than one sectionalizing switch or protective device will be
installed in each feeder section the following constraint is considered:
Xsf r + Yf f r ≤ 1 (5.16)
The distribution test network used in this chapter is the radial distribution network con-
nected at Bus 4 of the RBTS, consisting of 38 load points, 51 and 38 possible sectionaliz-
ing switch and protective device locations, respectively [109]. The modified test network
without TS is depicted in Fig. 5.2. The required reliability data, such as customer data,
component failure data, average and maximum load at each load point are presented in
[109]. In addition, all switches and CBs are assumed to be automatic and with a 10
minutes switching time. The capital investment and installation cost of a sectionalizing
switch and fuse are assumed to be $4700 and $470, respectively [72]. The annual operation
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 117
T M Cf = RCf + M Cf (5.17)
M Cf = M HC + AT M H (5.18)
N R × CCf
RCf = (5.19)
T
The load growth rate of the system is assumed to be 3%. Furthermore, the CDF data
are elicited from Table 3.3. Note that, the aggregate or average cost model is adopted in
this chapter.
The proposed formulation has been implemented in the GAMS environment and is solved
by using the branch-and-reduce optimization navigator (BARON) solver on an Intel(R)
Core(TM) i5-2500 CPU@3.30GHz with 8 GBs of RAM. Nowadays, the interest in using
GAMS as a language for the representation of mathematical programming problems has
increased. Indeed, the GAMS is served as a common interface for different solvers such as
the BARON. BARON is one of the GAMS solvers for the global solution of MINLP.
The traditional MINLP algorithm can converge under specific convexity assumptions,
while BARON implements deterministic global optimization algorithms based on the
branch-and-bond type. It can guarantee convergence under a finite time and global optima
under fairly general assumptions.
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 118
LP14 LP17
LP11 LP12 LP13 LP15 LP16
Sectionalizing
switch 25 30
CB 20 22 24 27 29
F3 19 21 23 26 28
Fuse
LP8 LP9 LP10
14 16 18
F2 13 15 17
1 3 5 7 10
F1
2 4 6 9 12
Substation at Bus 4
8 11
57 59 62 64 67
61 66
F6 50 52 54
51 53 55
45 47 49
Figure 5.2: Modified distribution test system (with 38 load points (LP), 51 sectionalizing
switches and 38 fuses).
5.5.3 Results
In order to evaluate the proposed optimization method on the test network, three case
studies are performed. In the Case 1, the network with seven mandatory CBs on the
substation at each feeder and no sectionalizing switch and fuse are considered. It is obvious
that in this case the ECOST has the highest value equal to k$140.905. In the Case 2,
CBs, sectionalizing switches, and fuses are installed in all possible locations, which is the
base case. In this case, the network has the highest reliability level, but this is not the
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 119
desired option for the utilities just because of the high cost. Thus the optimization on the
number and location of these devices might be a good option while keeping the reliability
of the network on a reasonable level. In the Case 3, the optimal number and location of
sectionalizing switches and fuses are evaluated. The results of the case studies are shown
in Table 5.2. It should be noted that the TSs and alternate supply paths are neglected in
all these three cases.
In the Case 3, the optimal solution is k$59.577, which is 31.5% less than the base case.
The optimal number of fuses is 38 and the proposed optimization technique suggests to
put the fuse in all possible locations because of their lower cost when compared to the
cost of reliability. However, the TC value is relatively high for this case as the load points
downstream of a fault cannot be supplied due to absence of an alterative supply to restore
power to the customers. The program execution time was less than 2 seconds, which is
very small for these cases.
In order to ascertain the applicability of the proposed formulation to real size systems,
a distribution network comprising 420 feeders is considered which has been produced by
replicating 60 times the distribution network connected to Bus 4 of the RBTS. In this case,
the optimal locations of protective and sectionalizing devices are the same of Case 3. The
optimal solution is achieved in approximately 330 seconds which shows the applicability
of the proposed formulation to the real size system. The optimization statistics for the
Cases 3, 4 and the real size test system are presented in Table 5.3.
The size of the network just increased the computational time and the solver did
not fail to obtain the solution. In fact, the way that the problem is formulated can
provide the opportunity to apply the proposed formulation to radial systems of any size.
This is because each feeder in the network can be treated separately with the proposed
formulation. Thus, by increasing the size of the network in terms of adding more feeders
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 120
to the network, the proposed formulation can handle the problem and just the running
time will increase.
In order to assess the effects of different parameters on the optimal number and location of
sectionalizing switches and protective devices, several sensitivity analyses are performed
on the test network. Furthermore, the reliability worth of network reinforcement indices
are proposed to investigate both utilities and customers benefit.
5.5.4.1 Effect of TS
RW T S = T C − T CT S (5.20)
where T CT S and T C are the network TC with and without TS, respectively. The results
of this reinforcement for different cases are shown in Table 5.2.
In the Case 1, due to the absence of sectionalizing switches and protective devices,
just the CB at the substation will act to remove the fault. Consequently, adding TS and
providing alternative supply for feeder cannot be useful to improve the reliability. In the
Case 2, 31.85% improvement is achieved by this reinforcement. Furthermore, the optimal
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 121
Case 3
Study Case
Without TS With TS
Number of Switches 4 13
17B,39B,46B,54B 5E,10E,15E,17B,26B,36B,39B
Location of Switches*
46B,48B,52B,54B,63B
Number of Fuses 38 38
Location of Fuses All possible location All possible location
* Every line section has two candidate locations for sectionalizing switch installation
which are shown as B for the beginning of a line section and E for the end of a line
section.
Location TC ECOST SC
of TS (k$/yr) (k$/yr) (k$/yr)
F1 56.269 50.404 4.0818
F2 57.749 52.564 3.4015
F3 56.207 50.342 4.0818
F4 56.199 51.014 3.4015
F5 58.072 53.567 2.7212
F6 57.499 52.314 3.4015
F7 57.758 51.892 4.0818
number and location of sectionalizing switches and fuses are presented in Table 5.4 for
Case 3.
Note that, in the Cases 2 and 3, despite of the significant improvement in TC, some
load points cannot benefit from this reinforcement because of the location of TS, protective
devices, and structure of the network. In order to assess the effects of TS’s location on
the optimal solution, a sensitivity analysis has been done on the test network. To do this,
a TS is added to the feeders of the network and the value of TC is calculated by the
proposed formulation. The results, presented in Table 5.5, demonstrate that the Feeder
F4 is preferable to put the first TS. It is worth noting that in the case of availability of
more than one TS for installation, for instance four TSs, the feeders F4, F3, F1 and F6
are preferred to put TSs, respectively. In this case, the value for TC is equal to k$50.162
while 13 sectionalizing switches and 38 fuses are suggested for the optimal solution. It can
be evidenced that adding more TS, allows reducing the objective function even if a higher
number of installed sectionalizing switches is proposed.
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 122
The following analysis has been carried out to evaluate the capability of the proposed
method to adapt to changes in feeder load profile or network topology to enhance the
reliability level of a distribution network, even when pre-existing sectionalizing switches
already exist. A constraint is defined to consider the location and number of previously
installed switches on the feeder. The relocation cost (RC) of switches is also added to the
SC formulation. The RC of a switch is assumed to be 10% of the capital investment cost.
The case of pre-existing switches (Case 5) is compared with the case of 12 available switches
for installation (Case 4, Fig. 5.3). In the Case 5, four previously installed switches at 5E,
10B, 39B and 63B locations are assumed. Fig. 5.4 shows the variation of the objective
function of the proposed formulation in both cases. Note that, the proposed optimal
number for fuses is 38.
The proposed method relocated three sectionalizing switches which already exist on
the network. However, the previously installed sectionalizing switches can directly affect
the utility cost of reliability (UCR). Therefore, by using the proposed formulation to
allocate and relocate sectionalizing switches, utilities can ensure that related expenditures
are made to achieve the higher reliability level in distribution networks. This sensitivity
analysis also verifies the effectiveness of the proposed formulation to take into account
the current locations of sectionalizing switches and protective devices which are really
important to reduce UCR and improve reliability.
LP14 LP17
LP11 LP12 LP13 LP15 LP16
Sectionalizing
switch 25 30
CB 20 22 24 27 29
F3 19 21 23 26 28
Fuse
LP8 LP9 LP10
14 16 18
F2 13 15 17
1 3 5 7 10
F1
2 4 6 9 12
Substation at Bus 4
8 11
57 59 62 64 67
61 66
F6 50 52 54
51 53 55
45 47 49
is reached. Beyond this point, the benefit of each additional sectionalizing switch is less
than its cost, and therefore introducing more switches does not reduce the TC effectively.
Note that, the proposed optimal number for fuses in this case is 38 due to the lowest price
of fuses in comparison with other devices.
Accordingly, the number of fuses is increased from 10 to 30, to indicate the effect of
fuse placement on the reliability of the network in terms of TC. Note that the number of
available switches is limited to 12. Changes in TC and ECOST as a function of the number
of available fuses are presented in Table 5.6. The results show the maximum available fuses
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 124
42.5
42
41.5
41
TC (k$/yr)
40.5
40
39.5
39
38.5
Case 4 Case 5
4
x 10
6
4
Cost ($/yr)
2
TC
1 ECOST
SC
0
5 10 15 20 25 30 35 40 45 50
Number of sectionalizing switches
installed in laterals according to the switches locations. For instance, in the case of having
20 fuses available for installation (Fig. 5.6), the proposed formulation placed all the 20
fuses on the laterals because of their lowest price to improve the reliability of the network.
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 125
LP14 LP17
LP11 LP12 LP13 LP15 LP16
Sectionalizing
switch 25 30
CB 20 22 24 27 29
F3 19 21 23 26 28
Fuse
LP8 LP9 LP10
14 16 18
F2 13 15 17
1 3 5 7 10
F1
2 4 6 9 12
Substation at Bus 4
8 11
57 59 62 64 67
61 66
F6 50 52 54
51 53 55
45 47 49
To further establish the effectiveness of the proposed formulation, the proposed location
of the first seven sectionalizing switches and fuses is investigated. The solution derived
from the proposed formulation shows that the customer types and load density at each
load point of the feeder are crucial parameters in distinguishing the optimal number and
location of switches in the network. Three different scenarios for switch and fuse placement
sequence are evaluated in this section.
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 126
Table 5.6: TC and ECOST vs. Number of Available Fuse for Installation
Number TC ECOST
Study Case Location of Switches
of Fuses ($/yr) ($/yr)
5E,10E,15E,23E,28E,36B
Case 6 10 50177.56 41544.57
41E,46B,48B,52E,60E,65E
5E,10E,15E,17B,23E,28E
Case 7 20 46340.91 37238.59
41E,46E,52E,54E,60E,65E
5E,10E,15E,23E,28E,36B
Case 8 30 43512.94 33941.30
41E,46E,52B,52E,60E,65E
In the first scenario, the sequence of switch placement on the network with TS for
all feeders and the optimal number of fuses is evaluated. The results are summarized in
Table 5.7. As shown, the location 41E of Feeder 4 is proposed for the placement of the first
switch. According to [109], Feeder 4 has the highest load level and contains two commercial
customers with comparatively high CDFs. Since the CDFs for commercial customers are
substantially greater than those of non-commercial customers, the proposed locations of
the first three switches are on feeders with higher loads and adjacent to commercial sectors.
The forth and fifth switches are proposed to be placed on feeders with grater loads and
adjacent to small-users sector. Furthermore, as the main load of Feeder 7 consists of
residential customers with a high load level, the 6th switch is proposed to be located in
this feeder. It is important to note that for the feeders with the same type of customers
(i.e., residential or small-user customer), the sectionalizing switches are proposed to be
allocated in the middle of feeders.
In the second scenario, the sequence of fuse placement on the test network with TS for
all feeders and 12 available sectionalizing switches is evaluated. The summarized results
in Table 5.7 indicate that the type and number of customers in feeders are important
parameters in determining the location of fuses in the network. For instance, the first
two fuses are placed on the laterals of Feeder 4 which has eight load points and consists
of two commercial load points. As mentioned before, because of a higher CDF value for
commercial customers, the first seven fuses are located on the laterals adjacent to the
commercial sectors.
In the third scenario, the sequence of placement of sectionalizing switches and fuses on
the test network with TS for all feeders is evaluated. To do this, the proposed formulation
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 127
is used to locate the first switch and fuse on the test network, simultaneously. Then, the
second switch and fuse are located while considering the presence of pre-installed first
switch and fuse in the network. This procedure is continued to locate seven switches and
fuses in the network. As shown in the results, the sequence of switch and fuse placement
is followed by the aforementioned scenario. The suggested locations for the switches and
fuses verifies the performance of the proposed formulation as the sequence of switch and
fuse placement meet the reliability constraint requirements.
Table 5.7: Sequence of Fuse and Switch Placement on the Test Network
According to the utility practices, the lateral taps can be protected by fuses or, in the
case of heavily loaded or long laterals, various types of protective devices can be installed
at the beginning section or at other points of the laterals. To cover such requirement
that might be imposed by the utilities, in proposed formulation, if the installation of a
protective device is mandatory/or not in a given section the corresponding variable can
be set to one/or zero. In general, a CB or a line recloser can be installed at the tap of
heavily loaded or long laterals. But due to the cost limitations, only a limited number of
these devices can be installed on the laterals and finding the optimal number of them is of
interest. To do this, the laterals consisting of Load Points 6, 7, 16, 17, 24, 25 and 38 are
considered as the heavily loaded laterals and are candidate to install CB or line recloser.
Reclosers have switching and protection functions, and are proper to handle both
temporary and permanent faults. The capital investment and installation cost of a recloser
is assumed to be $8 300 and the annual operation and maintenance cost of a recloser is
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 128
The results are shown in Fig. 5.7. The reclosers are placed on feeders with relatively
high load levels and consisting of the heavily loaded laterals. The TC value in this case
(Case 11) is $47 842, which is a great reduction on the utilities’ cost and meet their
practices. This is because the fuses can only perform an open-circuit function, and are not
able to clear the momentary faults by themselves. But, reclosers can prevent momentary
faults from becoming permanent interruptions by their trip/reclose function and as a result
can decrease the number of interrupted load points by their fault isolation.
To demonstrate the superiority of the proposed MINLP formulation and for comparison
reasons, the distribution feeder presented in [2] is selected. In [2], the problem is solved
by linear integer programming, where the nonlinear terms are translated by additional
constraints. The optimal number and locations of switches in [2] are taken and summarized
in Table 5.8. The TC associated with these number and locations is recalculated using
the same data with the proposed MINLP. It can be seen that the value for TC by the
proposed MINLP is less than that of [2], while 7 switches are suggested for the optimal
solution (Fig. 5.8). Also, the switch locations on each feeder by the proposed MINLP are
somehow close to the customers with the higher CDF. Furthermore, only one switch is
located on Feeder 2 with the lowest load levels.
Also, the number of available sectionalizing switches in the proposed MINLP is limited
to 4, and the results are shown in 5.8. The value of TC is less than that of [2] and confirms
that the proposed MINLP formulation is more effective to improve the reliability of power
distribution networks compared to MILP due to its linear approximation. Note that, based
on the analytical study in Chapter 3, the start point is selected for the proposed MINLP
as a primary solution to have a better performance.
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 129
LP14 LP17
LP11 LP12 LP13 LP15 LP16
Sectionalizing
switch 25 30
CB 20 22 24 27 29
F3 19 21 23 26 28
Fuse
LP8 LP9 LP10
14 16 18
F2 13 15 17
1 3 5 7 10
F1
2 4 6 9 12
Substation at Bus 4
8 11
57 59 62 64 67
61 66
F6 50 52 54
51 53 55
45 47 49
Sectionalizing
switch
CB TS
36 37 38 39 40 41 42 43 44 45 46 47 48
Fuse F4
Substation at Bus 5
LP20 LP21 LP22 LP23 LP24 LP25 LP26
TS
F3 25 26 27 28 29 30 31 32 33 34 35
Figure 5.8: Optimal location of sectionalizing switches by the proposed MINLP of [2]
test system.
5.6 Conclusion
This chapter has presented a mixed-integer nonlinear formulation to identify the types and
locations of protective devices and sectionalizing switches using value-based optimization.
A mathematical optimization methodology has been implemented to minimize the relia-
bility indices of a distribution network. To improve the continuity of service, the main
goal of the proposed formulation was to develop a more accurate model for the fault clear-
ance, separation and power restoration procedure by minimizing the total cost of reliability
while installing the minimum number of protective devices and switches. More accurate
modelling provided better solutions to meet the utilities practices.
Effectiveness of the proposed formulation has been validated by means of case studies
on Bus 4 of the modified RBTS. Numerical results and sensitivity analyses have confirmed
the efficient performance of this formulation. Furthermore, the effects of TS, pre-installed
devices, budget limitation and the importance of customer type on the optimal protec-
tive and switching devices placement were investigated through sensitivity analysis. The
Chapter 5. Optimal Type, Number and Location of Protective Devices in Distribution
Systems 131
benefits of the proposed optimization can be compared with the reliability worth to assist
decision makers in making proper planning and operating decisions.
Chapter 6
132
List of Symbols
Symbol Definition
νb Averaged value
eb Averaged value
S Correlation function
nI Dimension of the input layer
λdg Expected failure rate of a DG unit
µdg Expected repair rate of a DG unit
β Mean of the normal transformed distribution
fi (·) Nonlinear activation function
x Original data
νt Output of candidate unit
Pz Proportion of zero-valued data
α Power exponent
et Residual network output error
N Total number of training patterns
y Transformed value
σ2 Variance of the normal transformed distribution
133
Chapter 6. Reliability Analysis of Distribution Systems Considering Customer
interruption Cost Model 134
6.1 Introduction
Reliability worth evaluation is one of the most important functions in distribution network
planning and operation. The accuracy of the reliability worth can be affected directly by
the customer interruption cost model. The load points and system reliability indices can
change due to the different cost models. In this chapter, a cascade correlation neural
network is used to develop two interruption cost models including an average or aggregate
model and a probabilistic distribution model. A contingency based analytical technique
is adopted for the reliability worth analysis. In addition, the impact of introducing DG
units into the network is presented. The description of the proposed approach is tested by
evaluating the reliability worth of a radial distribution test network.
Incorporating cost analysis and quantitative reliability evaluation are used for the reliabil-
ity worth assessment. Also, the reliability worth analysis provides a value-based assessment
which reflects the integrated effects of interruption frequency and duration in distribution
systems. There are two different approaches to measure reliability worth. The first ap-
proach is a direct evaluation of reliability worth which is very difficult and sometimes
impossible to implement [130]. Thus, the monetary estimation is used to perform di-
rect assessment. Another approach is an indirect measure in which the interruption costs
are mostly used. A variety of methods have been introduced to evaluate the customer
impacts due to interruptions [21]. These methods can be divided, based on the method-
ological approach, into three extensive groups: analytical evaluation methods, case studies
of blackouts, and customer surveys [22].
The customer survey approach provides customer outage cost data that could be mod-
eled as the customer damage function (CDF) [131], [23]. This model determines the
aggregate or average cost of interruption for each customer sector as a function of dura-
tion. Consequently, the basic cost models to implement reliability worth analysis can be
derived by CDF. The main disadvantage of the aggregate or average cost model (AAM)
is that the dispersed nature of data within a specified customer group is neglected. The
Chapter 6. Reliability Analysis of Distribution Systems Considering Customer
interruption Cost Model 135
dispersion of the customer interruption cost data is important information and can have
a significant effect on the accuracy of the reliability assessment. The dispersed nature of
customer interruption costs at specified failure duration has been designated as the proba-
bility distribution cost model (PDM) [24]. The results presented in [24] have demonstrated
that the PDM has a considerable effect on predicting the expected customer interruption
cost.
Recently, artificial neural networks have widely gained more attention for many power
applications. One of the most important challenges in the application of neural networks
is the definition of a proper network architecture. The radial basis function (RBF)[18],
multilayer perceptron (MLP) [132] and self-organized map (SOM) [133] architectures have
been wieldy used due to approximation ability, flexibility of structure and various training
algorithms. However, recent studies have shown that a large number of hidden layers is
used in these network architectures and they are less powerful than the cascade correlation
(CC) architecture [134], [135]. The CC is a feed-forward and supervised learning algorithm
for artificial neural networks [136]. Instead of algorithms that deal with just fixed-size
networks, the CC neural network (CCNN) begins from the simplest structure without
hidden units. Then, the algorithm automatically trains and installs new hidden units to
the network one by one to create a multi layer structure until a given performance criterion
is met. The CCNN algorithm is used in this chapter because of its advantages such as: 1)
it learns very quickly since only weights of a single hidden unit are trained at any time;
2) the network has the ability to determine its own size and topology; 3) in the case of
changes in the training set, the network can retain the structures it has built, therefore
there is no need for back propagation of error signals.
However, the performance of the CCNN depends upon the training procedure of the
weights feeding into a new hidden unit and the weights that connect the input units and
all the hidden units to the output units. For each new hidden unit, the input weights
are trained to maximize the magnitude of the correlation between the output of new
hidden unit and the residual error signal that needs to be eliminated. However, maximal
error reduction cannot be achieved by this objective function and may slow down the
algorithm convergence. Also, after adding each new hidden unit, it is necessary to train
the weights that connect input and hidden units to the output units repeatedly, which is
Chapter 6. Reliability Analysis of Distribution Systems Considering Customer
interruption Cost Model 136
time consuming. This shortcoming can be overcome by reformulating the training problem.
An intelligent training procedure based on the orthogonal least squares (OLS) has been
derived in this chapter, which preserves the advantages of the new objective function for
input training. The method provides a maximal sum of squared errors (SSE) reduction
when the new hidden unit is added. Meanwhile, there is no need to train the weights
that are related to the output units repeatedly [134]. The OLS training procedure chooses
the appropriate number of hidden units, greatly reducing the network size and providing
better performance.
The main contribution is the development of the CCNN to construct AAM and PDM
cost models for reliability worth assessment in distribution networks. To demonstrate
the superiority of the proposed CCNN, the results are compared with two of the existing
neural networks. In this cahpter, a contingency-based analytical technique is adopted for
the reliability worth assessment with two customer cost models based on CCNN. However,
the effect of DG units is an important one and has not been considered in [137] and
[138]. Therefore, in this chapter, based on the islanding operation of DG, a probabilistic
formulation to calculate the reliability indices of the network is also reported. In addition,
a two-state model for DGs and a time varying load model are considered, which can
provide a more accurate analysis and have not been considered in the cited references.
Furthermore, to illustrate the primary protection system fails to operate, which has an
important effect on the results, the probability of switch operation is also considered in
the reliability analysis.
6.3 CCNN
6.3.1 Fundamentals
The CCNN is one of the efficient constructive algorithm that has a dynamic architecture.
The CCNN has a topology of three layers but initially consists of the simplest structure
in which the input layers are directly connected to the output layers as dictated by the
problem to be solved. The CCNN architecture is presented in Fig. 6.1. All connections
from input to output layers have adjustable weights, which can be trained to minimize an
Chapter 6. Reliability Analysis of Distribution Systems Considering Customer
interruption Cost Model 137
Second Hidden
Units
First Hidden
Units
Input Units
error such as the SSE. The output layer may just produce a linear sum of their weighted
inputs, which is not good for a complex problem and nonlinear activation functions may
be required.
The training procedure starts with no hidden units by using quickprop algorithm [139],
which converges quickly to achieve a sufficiently low error rate. If this fails, it is necessary
to add new hidden units to reduce or eliminate the residual error. A pool of candidate
units which can receive trainable input connections from the input units and all the pre-
existing hidden units is used instead of just one unit. The output of the new candidate
unit is not yet connected to the active network. Then, the input weights of candidate unit
is trained to maximize S, which is the correlation function between the output value of
candidate unit ν and the residual network output error e:
N
X
S=| (νt − νb)(et − eb)| (6.1)
t=1
where N is the number of training patterns, νt and et are the output of the candidate unit
and the residual error of the network for pattern t before adding the candidate unit. νb
Chapter 6. Reliability Analysis of Distribution Systems Considering Customer
interruption Cost Model 138
Once all the candidate units with the largest correlation value have been trained, the
most successful candidate is chosen to connect to the output layer. Once a new hidden
unit is added, all the output unit weights are then trained.
As mentioned before in this chapter, initially, the CCNN consists of only input and output
layers which are directly connected together with an adjustable weight. This framework
is considered as the basis units, which can be expressed as:
the ith input unit 1 ≤ i ≤ nI
the (i − n )th hidden unit i ≥ nI
I
where nI refers to the dimension of the input layer. The full mode is defined by considering
a training set of N samples {d(t), x(t)}N
t=1 , where d(t)∈R is the desire network output
corresponding to the input x(t)∈RnI . The output of the ith basis unit is derived by:
xt,i
1 ≤ i ≤ nI
νt,i = i−1
P
fi (
νt,j wj,i ) i ≥ nI
j=1
(6.2)
where fi (·) is defined as a nonlinear activation function. The weights that connect the
input units and pre-exsiting hidden units to the ith basis units are depicted by wj,i .
The output of the network with mapping of fi : RnI → RnO with i basis units, is a
linear activation function defined by:
i
X
dt,i = νt,j θj t = 1, ..., N (6.3)
j=1
Chapter 6. Reliability Analysis of Distribution Systems Considering Customer
interruption Cost Model 139
where θj are the weights that connect basis units to the output.
The OLS procedure can be implemented by introducing an error term et , and (6.3) can
be rewritten as follows:
i
X
dt,i = νt,j θj + et,i t = 1, ..., N (6.4)
j=1
where dt,i is the desired value in the output layer. Equation (6.4) can be expressed as a
matrix form
D = νΘ + E (6.5)
The least square principle can be used to solve the matrix parameter Θ. An orthogonal
decomposition of the regression matrix ν can be presented as follow:
ν = WA (6.6)
Since the error matrix E is orthogonal to W, the aim is to minimize the SSE described
as follows:
Chapter 6. Reliability Analysis of Distribution Systems Considering Customer
interruption Cost Model 140
N
X
Ei = e2t,i = eTi ei (6.8)
t=1
i
X (WT D)2
Ei = DT D − k
i≥1 (6.9)
k=1
WkT Wk
According to the cascade network training procedure, since the trained input weights
are frozen, the output of existing hidden units would be fixed while the network becomes
large. Thus, the error reduction ratio due to introducing the ith basis unit to the network
can be derived from (6.9):
(WkT D)2
[err]i = Ei−1 − Ei = i≥1 (6.10)
WkT Wk
where Ei−1 refers to the condition before adding the ith basis unit. By using (6.10),
contribution of a candidate unit on the SSE reduction can be analysed without solving
the least square problem explicitly. Equation (6.10) is used as an objective function to
train the input layer in which, concluded the SSE reduction, can be maximized when a
new hidden unit is introduced [134].
In order to find the output weight, the following equation can be derived from (6.7):
AΘ = λ (6.11)
where
WiT D
λi = (6.12)
WiT W
By using (6.11), the output weights can be solved using back substitution. As men-
tioned before, the output weights are not needed when the input training procedure is
running, thus it is not necessary to fulfill the output training until all hidden units are
introduced. The outline of the implemented method is presented in Fig. 6.2.
Chapter 6. Reliability Analysis of Distribution Systems Considering Customer
interruption Cost Model 141
Start
The application of CCNN to propose two interruption cost models including AAM and
PDM is investigated in this chapter.
6.4.1 AAM
Postal surveys [140] have been implemented to appraise the customer interruption cost
(CIC) for different customer sectors. The survey data have been assessed to extract the
sector CDF related to each group of customers which is presented in Table 6.2. This Table
describes the interruption cost for the given type of customer as a function of interruption
duration. It should be noted that a log-log interpolation of the cost data is used where
the interruption duration has a value between separate times. For values greater than
480 min, a linear extrapolation with the same slope from 240 min was implemented to
compute the interruption cost.
According to the architecture of cascade networks, the three inputs vector related to
the proposed CCNN for AAM are defined as X1 , X2 , X3 . The first two inputs imply the
Chapter 6. Reliability Analysis of Distribution Systems Considering Customer
interruption Cost Model 142
type of customers and X3 emphasizes the interruption duration. In the output layer there
is one vector that indicates the sector interruption cost for each customer. The values
sampled with 10 minutes duration are chosen as the training data for the proposed CCNN
[138].
6.4.2 PDM
The data collected from the AAM can be used to evaluate the reliability cost through
the CDF method in which a measure of the tendency of customer interruption data is
provided. It should be noted that the CDF method cannot reflect the dispersed nature of
the actual interruption cost. Interruption cost evaluation demonstrates that the monetary
values expose a large deviation and, in some cases, the standard deviation is more than four
times the mean value [137], [24]. Having taken into consideration the dispersed customer
interruption cost data in the analysis, the actual data need to be transferred into a flexible
mathematical model.
A normality transformation has been used to represent the full range of interruption
cost data. According to this systematic procedure, the customer interruption cost data can
be transformed into the normal distribution in which the PDM is developed to represent
the data [137]. In this approach, the data collected from the survey for the specified dura-
tion are transformed into a group of data, which are illustrated by a normal distribution
using the normality transformation. The following transformation is selected for the work
presented in [24]:
xα −1
if α 6= 0
α
y= (6.13)
log(x) if α = 0
Chapter 6. Reliability Analysis of Distribution Systems Considering Customer
interruption Cost Model 143
Duration(min) α β σ2 Pz (%)
1 -0.0488 0.3352 3.8770 0.3488
20 -0.0605 1.0487 2.7866 0.1513
60 -0.0707 1.6327 2.3443 0.0613
240 -0.0387 2.8272 2.3620 0.0047
480 -0.0020 3.6939 2.9880 0.0047
Duration(min) α β σ2 Pz (%)
20 -0.2207 -5.6618 4.8689 0.3295
60 -0.1828 -2.7329 2.8790 0.0973
240 -0.0105 0.2886 1.6551 0.0265
480 -0.0160 1.1345 1.5725 0.0426
where x is the original data, α is the power exponent and y is the transformed value. This
family of equations has two limitations: (i) it is applicable only to continuous variables
and (ii) it does not perform well when applied to zero-valued data. The zero-valued
customer outage costs, which were extracted from the specific duration data, can be treated
separately to satisfy the above mentioned constraints. An iterative process to distinguish
the value of α is employed to analyze the normality of the remaining data.
The distributed nature of the transformed interruption cost data for a particular cus-
tomer sector and a specific outage duration can be determined by four parameters: the
normality power transformation factor α, the mean of the normal transformed distribution
β, the variance of the normal transformed distribution σ 2 and the proportion of zero-valued
data Pz . These parameters for industrial and residential customer sectors are presented
in Table 6.3 and Table 6.4. Note that, the data collected by the survey was related for
two types of customers as industrial and residential. The other types on customers are
not included in this survey and they are not discussed in this dissertation.
In this chapter, regression analysis is used to predict the distribution pattern for in-
termediate durations. The relation between the studied duration (d(min)) and each of
the four parameters is described by the following equations which are derived by the least
square method [137]:
Industrial customers:
Chapter 6. Reliability Analysis of Distribution Systems Considering Customer
interruption Cost Model 144
0.3333 − 0.1346log(d)
d < 4 hours
Pz = (6.17)
d ≥ 4 hours
0.0047
Residential customers:
Pz = 0.8771[log(d)]−3.7322 (6.21)
where the log base is 10. To determine the parameters of the normal distribution and the
zero cost, (6.14)-(6.21) can be used as cost models.
The proposed CCNN for PDM contains three inputs such as the AAM. The output
layer consists of four parameters of PDM (i.e. α, β, σ 2 and Pz ). The values sampled from
(6.14)-(6.21) with 10 minutes duration are chosen as the training data for the proposed
CCNN.
Chapter 6. Reliability Analysis of Distribution Systems Considering Customer
interruption Cost Model 145
It should be noted that the customer interruption cost cannot be directly described
by the PDM data. The data of the PDM have to be transferred back to the original
form to generate the actual customer interruption cost. The procedure to calculate the
customer interruption cost using PDM is more complex compared to the CDF method.
The proposed CCNN can be used to obtain four parameters of PDM for different types of
customers.
The actual customer outage cost x ($/kW ) corresponding to the parameters, which is
used to assess the reliability worth indices, can be calculated using the following inverse
transformation:
(1 + α · y)1/α
if α 6= 0
x= (6.22)
log −1 (y)
if α = 0
The model for components in distribution systems related to this analysis are described
in this section.
failures, respectively. The rate of active and passive failures are displayed by λa and λp ,
respectively. rs and rp are defined as the time spend in the switching and repair states.
The failure rates are assumed to be constant for normal weather conditions and the adverse
and major storm disaster conditions are ignored.
The probability of operational failure (POF) is the conditional probability that a device
fails to operate. It is associated with the protective and switching devices and is assumed
to be 10% in this study.
Normal
State
Repair Switching
Repair Switching
State State
(a)
Normal
Failure Repair
Repair
(b)
1 / rp Normal
Pasive a Failure
Active
Failure
p
Repair Switching
1 / rs
(c)
Figure 6.3: State diagrams for components in distribution systems. (a) three-state
model, (b) two-state model, and (c) combined state model.
Chapter 6. Reliability Analysis of Distribution Systems Considering Customer
interruption Cost Model 148
6.5.3 DG Model
dg
DG Up DG Down
dg
Figure 6.4: Two-state model for DG units.
The probability of availability and unavailability of the two-state model for DG can be
expressed by:
µdg
Pavailable = (6.23)
λdg + µdg
λdg
Punavailable = (6.24)
λdg + µdg
and a quantification of the value of delivering it. Typically, when the supply to a load is
interrupted, there is a cost associated with the loss of such supply.
The load point and system indices that sufficiently represent the cost of reliability are
the ECOST (5.1) and IEAR [7]. The IEAR can be calculated as follows:
ECOST
IEAR = ($/kW h) (6.25)
EN S
Island formation occurs when a part of the distribution system, that includes one or more
DG units, becomes electrically isolated from the rest of the system after a fault clearance
procedure, but continues to be energized by the DG units in order to supply the loads
of the healthy parts of the system. In order to maximize the benefits of utilizing DG
in the distribution networks, it is important to consider the island operation possibility
in distribution networks. In this chapter, the proposed island operation introduced in
Chapter 3, is adopted in the case of LOM or outage in the distribution system component.
The modified distribution test system used in this chapter is the radial distribution system
at Bus 2 of the RBTS, consisting of 22 load points and 10 sectionalizing switches [109].
The modified test system without normally open switches is depicted in Fig. 6.5. Required
reliability data, such as customer data, component failure data, average and maximum load
at each load point, etc., are presented in [109]. Load Points 1-7 and 10-19 are considered as
residential customers, and Load Points 8-9 and 20-22 consist of industrial customers [137].
In addition, all switches are assumed to be automatic and take 10 minutes for switching
time. The data for AAM and PDM cost model has been collected by the survey [140]. In
order to illustrate that the primary protection system fails to operate, it is assumed that
the switches operate with the probability of 0.9. In this situation, the reliability indices
Chapter 6. Reliability Analysis of Distribution Systems Considering Customer
interruption Cost Model 150
11kV
28 31 36
F4 26 27 30 32 33 34 35
29
F3
16 18 21 24
17 19 22 25
20 23
LP8 LP9
1 4 7 10
2 5 8 11
F1 3 9
6
should be modified. The contribution to the load point interruption frequency (LPIF) can
be assessed using the concept of expectation as follows:
1
10
RBFNN
Sum−Squared Error
CCNN
BP
0
10
0 2 4 6 8 10 12 14 16 18 20 22 24 25
Epoch
Figure 6.6: RMS error convergence diagram for BP, RBFNN and CCNN.
The CCNN for PDM is compared with the back-propagation (BP) and RBFNN [138].
The BP network consists of one hidden layer with 25 neurons that converges slowly. The
sum-squared error value decreased from 1.084 to 0.689 after around 100 000 epochs, 1117
seconds on an Intel(R) Core(TM) i5-2500 CPU@3.30GHz with 8 GBs of random-access
memory (RAM). The number of neurons in the hidden layer of RBFNN is determined
by the OLS learning procedure [138]. The convergence rate for the RBFNN with 25
epochs was 1.8 s. Fig. 6.6 demonstrates the root-mean square (RMS) error convergency
for the three networks. The results indicated that the performance of the CCNN is better
than that of the BP and RBF networks in terms of training time and accuracy, which is
convenient for real-time applications. Note that the convergence rate for the CCNN with
12 epochs was 0.9 s.
In order to demonstrate the applicability of the CCNN method to real size systems,
a distribution network comprising 240 feeders is considered, which has been produced by
replicating the distribution network connected to Bus 2 of the RBTS 60 times. The RMS
value is assumed to be 0.001. The convergence rate for the RBFNN with 71 epochs is 10.3
s, while the CCNN generated smaller network with 23 epochs required 2.6s convergence
rate.
Chapter 6. Reliability Analysis of Distribution Systems Considering Customer
interruption Cost Model 152
The proposed CCNN for AAM and PDM, based on the survey data [140] for two types of
customers (i.e., industrial and residential), is implemented to assess the load points and
system reliability indices, and the results are presented in Tables 6.6 and 6.7. It can be
seen from Table 6.6 that the ECOST of Load Point 7 increases by 127% from $222.4853
to $505.2002 per year, and the ECOST of Load Point 8 increased by 401% from $2633.389
to $13203.1. However, the results indicate that the appraised CIC increases significantly
by considering the PDM, in which the PDM model provides a more realistic picture about
how cost due to power failure can be accurately reflected. In contrast, considering the
AAM model without the variation in the survey data can result in low estimated CIC.
It should be noted that the customer type and variation in the cost data can affect
the evaluation of the CIC. For example, the obtained CIC for the industrial customers by
PDM has a greater increase compared to residential CIC. This is due to the fact that the
interruption cost of industrial customers is bigger than the residential one and the related
cost data have a greater variation. It can be concluded that the PDM cost model can
significantly affect the ECOST and IEAR, while the AAM model tends to underestimate
the CIC.
6.7.3 DG Impact
The reliability worth of introducing DG (RWDG) into the distribution system has an
important role in power system planing, design and expansion. If the RWDG is large
enough or equals the investment in the augmentation, both customers and utilities will
benefit. The RWDG can be expressed using the following equation:
RWDG ECOST(k$/year)
Feeder
(k$/year) Case1 Case2
F1 0.3917 1.2457 0.854
F2 0.6693 4.4123 3.743
F3 0.4024 1.1184 0.716
F4 2.5865 7.2975 4.711
Total 4.0500 14.074 10.024
RWDG ECOST(k$/year)
Feeder
(k$/year) Case1 Case2
F1 0.9213 2.8971 1.9758
F2 3.3574 20.997 17.639
F3 0.9418 2.6233 1.6815
F4 8.3819 23.638 15.256
Total 13.602 50.155 36.602
where ECOSTDG and ECOST are the system or load point interruption costs with and
without DG, respectively.
The reliability worth of the distribution test system (Fig. 6.5) in presence of DG is
illustrated here. The distribution test system is first modified into a basic structure (Case
1), in which there is no DG. In Case 2, four DG units (DG1-DG4) are connected to the
Feeders F1-F4, respectively. The two cases are analysed and compared to determine the
corresponding reliability worth of the stated reinforcement.
The system and feeder ECOST and RWDG using the AAM and PDM for Cases 1
and 2 are shown in Tables 6.8 and 6.9, respectively. The system RWDG of adding DG at
Feeder 1 is k$0.3917 per year using the AAM and k$0.9213 using the PDM. The AAM
clearly underestimates the reliability worth of adding DG into the system. On the other
hand, ignoring the disperse nature of the collected data can led to imprecise results. This
analysis can prove that the PDM model would be more effective to calculate the reliability
indices.
Fig. 6.7 shows the comparison of the load point RWDG using the two cost models. It
can be seen that the RWDG values obtained by the PDM are greater than those obtained
by AAM. As expected, using the PDM result in a grater value for RWDG compared to
Chapter 6. Reliability Analysis of Distribution Systems Considering Customer
interruption Cost Model 155
3.5
2.5
RWDG (k$/yr)
2
1.5 AAM
PDM
1
0.5
LP10
LP11
LP12
LP13
LP14
LP15
LP16
LP17
LP18
LP19
LP20
LP21
LP22
LP1
LP2
LP3
LP4
LP5
LP6
LP7
LP8
LP9Load Points
AAM for Feeders 1-4. Note that, the RWDG at some load points might be zero because of
the location of protective devices and structure of the system. Therefore, according to the
location of DG units into the system, some load points benefit from this reinforcement.
6.8 Conclusion
In this chapter, a CCNN has been developed to construct AAM and PDM for reliability
worth analysis in distribution systems. The OLS training procedure provides a great
reduction on the network size with a good speed. The CCNN has also been compared
with BP and RBFNN, and the results show that the CCNN has a better performance and
accuracy.
An analytical technique has been used to evaluate the interruption cost of individual
load point and system. Different models are considered for the elements in the test network
to provide a more accurate assessment.
the AAM. In fact, the PDM can result in a more realistic and effective assessment for the
reliability worth of network reinforcement.
Chapter 7
157
List of Symbols
Symbol Definition
Ljtkf r Annual average load of type k at Load Point j of Feeder f r at Year t
γ Annual load increase rate
CCc CB capital investment cost
ICc CB installation cost
CCf Fuse capital investment cost
ICf Fuse installation cost
Ctemp Interruption cost ($/kW) of temporary faults
CCs Sectionalizing switch capital investment cost
ICTijtkf r Interruption cost of Load Point j of type k due to temporary outage in Section
i of Feeder f r at Year t
ICs Sectionalizing switch installation cost
M Cs,t Sectionalizing switch annual operation and maintenance cost
T Sectionalizing switch and fuse life period
M Cc,t Total maintenance cost of a CB
M Cf,t Total maintenance cost of a fuse
NCT Total number of customer types
NLP f Total number of load points in Feeder f
NLP Total number of load points
Nac Total number of available CB for installation
Naf Total number of available fuses for installation
Nas Total number of available switches for installation
Nc Total number of installed CBs
158
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 159
7.1 Introduction
A more comprehensive power distribution system with higher efficiency, reliability and
power quality can be designed if utilities install new and advanced protective equipment to
satisfy the growing demand for service quality. Power distribution systems are normally
equipped with protective devices such as sectionalizing switches, circuit breakers, fuses
and reclosers, which are used to isolate faulted parts of the network and recover the power
supply from electrical faults. The role of these devices to improve the reliability of the
system is considerable. They can reduce the outage duration and failure rate, which
affect the load point, thus reducing substantially the customer interruption costs (CIC).
Therefore, utilizing DAS in terms of installing automatic and remotely control switches is
recognized as one of the ways to increase the reliability of the entire distribution system.
In fact, One of the functions of DASs is to determine the placement of remote control
sectionalizing switches and protective devices to isolate a fault and restore service to
the customers. However, the extend to which such improvement is delivered should be
quantified, and this will help to numerically evaluate the installation of such equipment.
Moreover, it is important to objectively evaluate the relative benefits of installing specific
types of protective devices and switches.
The identification of location and type of sectionalizing switches and protective de-
vices in distribution systems is a concern of power engineers with regards to reliability
assessment. Several types of protective devices are studied for reliability evaluation in
[71]. The authors quantified that the reliability improvement can be achieved by us-
ing each of these devices, as well as their combination. They used a reliability analysis
program designed to determine the methods for improving reliability of the distribution
system. Reference [143] discusses the challenges and strategies for promoting reliabil-
ity of the system when reclosers, sectionalisers and other switching devices are applied
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 162
in distribution feeders. More recently, technical papers have been published on opti-
mum allocation of sectionalizing switches and protective devices in distribution systems
[5, 9, 23, 72, 117, 118, 121, 123, 124, 144]. The factors such as optimum number, location
and type of such protective devices are important to improve the reliability of distribution
networks by minimizing the impact of the interruption.
Reliability worth evaluation is one of the most important functions in power system
planning and operations [145], which provides an opportunity to estimate the customer
outage costs as a result of a power failure. As mentioned in Chapter 6, the accuracy of
the reliability analysis can be affected directly by the customer interruption cost model.
The customer interruption cost model plays a key role in assessing the reliability of the
distribution system and provides an indirect measure of reliability worth. For instance,
the customer outage cost data obtained in postal surveys could be analyzed and sum-
marized in the sector customer damage function (SCDF) [140]. The basic cost models
for implementing reliability worth analysis is provided by SCDF. The proper cost model,
such as the PDM, which presents the dispersed nature of customer interruption costs at
specified failure duration, is used to evaluate the accuracy and validity of reliability on
the distribution systems. Therefore, by implementing the reliability worth analysis the
decision makers will be able to consider the value of improvement tasks on the system to
get a higher level of reliability.
Until recently, the grows of DG has been limited by economic reasons [146]. DG is nor-
mally defined as small generation units (<15MW) installed in distribution systems [147],
which includes traditional and renewable power generation sources. DG can affect the en-
tire system and particularly distribution systems, for instance, by providing emergency for
back-up supply to enhance the reliability and stability of the power supply at customer’s
site; changing the power flow distribution in the system and so on. However, how much
contribution the DG can provide to distribution system reliability is the very significant
issue that power utilities most care about. DG units can potentially contribute to improve
the reliability of distribution networks by reducing interruption duration and restoration
time after a given fault. However, such improvement depends on the operation mode of
the DG units, such as islanding mode, while the protection requirements are met.
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 163
The main contribution of this chapter is to propose an MINLP formulation to solve the
optimal allocation of sectionalizing switches and protective devices problem in presence
of DG units while considering the PDM cost model for the customer interruption. The
proposed method allows minimizing the cost of reliability while limiting the number of
installed switches and protective devices. The PDM cost model is based on a CCNN that
takes into account the costs associated with duration of interruption of customers in the
network. Furthermore, in this chapter, the cost of interruption due to the temporary faults
is also considered in the proposed MINLP formulation.
It is identified through the recent approaches that the costs incurred by the customers
due to power failure and the system costs are affected by the reliability. This means that
the reliability should be subjected to a more precise economic analysis. Incorporation of
cost analysis and quantitative reliability evaluation are opportunities provided by reliabil-
ity worth assessment. The major aspect is defined to evaluate the worth of the system
reliability in order to compare it with the costs of obtaining this level of reliability [7].
There are two basic approaches to conduct system reliability worth studies. The first
approach is the direct evaluation of reliability worth, which is very difficult and sometimes
impossible to implement. Thus, monetary estimation is used in order to perform direct
assessment. The second approach is the indirect measure, which interruption costs are
most often used to implement this method. Recently, a variety of methods have been
introduced to evaluate the customer impacts due to interruptions. These methods can
be divided, based on the methodological approach, into three extensive groups: various
indirect analytical evaluations, case studies of blackouts, and customer surveys. While
a single approach has not been universally adopted, utilities appear to favor customer
surveys as the means to determine specific information for their purposes [22].
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 164
A standard industrial classification (SIC) procedure was used to divide the customers
into seven group such as: large user, industrial, commercial, agriculture, residential, gov-
ernment and office categories. A precise approach to assess the impacts of customer service
interruptions due to failures in electric energy supply is provided by customer surveys. In
order to calculate the sector interruption cost as a function of the interruption duration, it
is necessary to collect the data from these surveys then use them to formulate SCDF. The
SCDF can be aggregated to create the composite customer damage functions (CCDFs),
which measure the cost associated with power interruptions as a function of the interrup-
tion duration for the customer mix at the bus or for the entire system.
Reliable distribution networks try to minimize the impact of failures to the customers
by utilizing more switches and protective devices and minimizing the number of customers
affected by protective device operations. Due to cost constrains, only a limited number of
switches and protective devices can be installed on feeders and laterals. However, types
and locations of these devices vary depending on utility practices and network topology.
Sectionalizing switches can be located at both sides of each line sections, and protective
equipment is usually located at the beginning of line sections on the feeders or laterals.
Operation of distribution networks can be evaluated according to the type, number and
location of switches and protective devices. For instance, when a fault occurs in the
Line Section 3 (LS3) of the network shown in Fig. 7.1(a), the following steps should be
performed:
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 165
Step 1 (fault clearance): The fault clearance functions open the protection breaker CB1
(Fig. 7.1(b)). Therefore, during the clearance of this fault, all load points are interrupted.
Step 3 (power restoration): The following switch actions are required to restore power
to the two separate restorable areas: 1) Separation switch S1 is remote-controlled and
has a switching time of 1 minute. Power to Load Point 1 is restored by (re)closing the
protection breaker CB1 which is also remote controlled. The LP1 is therefore restored in
1 minute. 2) Power to LP4 and LP5 is restored by closing the TS. Because the TS has an
actuation time of 30 minutes, loads 4 and 5 are restored in 0.5 hours. The network is now
in the post-fault condition as illustrated in Fig. 7.1(d).
However, LP2 and LP3 experienced the outage duration equal to the repair time of
LS3 which is much longer than the switching time.
It is worth to note that, when a fault occurs in the Lateral Section 2 (L2), as there is no
protective device in this lateral, the previous switching actions will be followed to clear this
fault. Thus, additional installation of protective devices on the lateral or sectionalizing
switches on the main feeder will increase the reliability of the network.
Consider the fully automated network with protective devices shown in Fig. 7.2. When
a fault occurs in LS3, all the load points can be restored after isolation of the faulty section
by the sectionalizing switches. Furthermore, the outage duration of the load points was
equal to the switching time of CB1 and the sectionalizing switches. In the case of fault in
L2, the presence of protective devices such as a fuse or CB will lead to isolate the fault
without interrupting other parts of the system. In this case, just LP2 experiences the
outage equal to the repair time of L2.
Although LP2 and LP3 would experience a shorter outage time in Fig. 7.2 compared
to Fig. 7.1, the switching and protective device costs in the network shown in Fig. 7.2
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 166
are much higher and the necessity of doing optimization is more important to balance the
cost of reliability and switching devices.
A variety of indices are defined in [7] to evaluate the reliability of distribution networks.
The basic indices that normally provide service reliability data from an individual customer
viewpoint are called single load point indices. These are include the load point average
failure rate (λj ), average outage duration (rj ), and average annual outage duration (Uj ).
The most frequently used customer oriented indices can be calculated using these three
basic indices, given as SAIFI, SAIDI, ENS and LPENS. These indices are not sufficient to
represent the cost of reliability. The LPIC and ECOST indices can be used to associate
reliability with customer costs since they accounts for the effects of the interruption du-
ration, failure rate of equipment, load variations, customer types and customer damage
functions [7]. The contingency-based analytical method, which is employed to calculate
the ECOST in distribution networks for a typical feeder, is well described in [117]. The
above mentioned indices can be calculated as follows.
Nq
X
λj = λi (7.1)
i=1
Nq
X
Uj = λi ri (7.2)
i=1
Nq
P
λi ri
Uj
rj = = i=1
Nq
(7.3)
λj P
λi
i=1
NP
LP
λj Nj
T otal number of customers interrupted j=1
SAIF I = = NP
(7.4)
T otal number of customers served LP
Nj
j=1
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 167
Sub1
Sub2
LS1 S1 LS2 LS3 LS4 S6 LS5 S8 LS6
CB1 TS
L1 L2 L3 L4 L5
Sub1
Sub2
LS1 S1 LS2 LS3 LS4 S6 LS5 S8 LS6
CB1 TS
L1 L2 L3 L4 L5
Sub1
Sub2
LS1 S1 LS2 LS3 LS4 S6 LS5 S8 LS6
CB1 TS
L1 L2 L3 L4 L5
Sub1
Sub2
LS1 S1 LS2 LS3 LS4 S6 LS5 S8 LS6
CB1 TS
L1 L2 L3 L4 L5
Figure 7.1: Sample of radial distribution network and fault restoration steps. (a) Fault
occurred in LS3, (b) fault clearance step, (c) fault separation step, and (d) fault restoration
step.
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 168
Sub1
Sub2
LS1 S1 LS2 S2 S3 LS3 S4 S5 LS4 S6 S7 LS5 S8 LS6
CB1 TS
L1 L2 L3 L4 L5
Sub1
Sub2
LS1 S1 LS2 S2 S3 LS3 S4 S5 LS4 S6 S7 LS5 S8 LS6
CB1 TS
L1 L2 L3 L4 L5
Sub1
Sub2
LS1 S1 LS2 S2 S3 LS3 S4 S5 LS4 S6 S7 LS5 S8 LS6
CB1 TS
L1 L2 L3 L4 L5
Sub1
Sub2
LS1 S1 LS2 S2 S3 LS3 S4 S5 LS4 S6 S7 LS5 S8 LS6
CB1 TS
L1 L2 L3 L4 L5
Figure 7.2: Sample of fully automated radial distribution network and fault restoration
steps. (a) Fault occurred in LS3, (b) fault clearance step, (c) fault separation step, and
(d) fault restoration step.
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 169
NP
LP
Uj Nj
Sum of all customers interruption duration j=1
SAIDI = = NP
(7.5)
T otal number of customers served LP
Nj
j=1
LP EN Sj = Uj × (pd
cj + ps
cj ) (7.6)
LP ICj = Uj × (pd
cj + ps
cj ) (7.7)
N
XLP
EN S = LP EN Sj (7.8)
j=1
Nq NLP NCT
X XX
ECOST = λi × CDFijk (rij ) × Ljk (7.9)
i=1 j=1 k=1
where λi is the average failure rate of the distribution network equipment i, CDFijk (rij )
is the customers damage function that depends on rij , rij being the failure duration of
jth load point, and Ljk is the average load of kth type customer located at the jth load
point. The LP ICj,i is the average interruption cost for load point j and contingency case i,
considering the load point interruption costs function and the assessed distribution of the
durations of the interruptions at this load point for contingency case i. The interruption
costs are calculated differently for different cost functions. All cost functions express
the costs as a function of the interruption duration. For cost functions expressed in
money per interrupted customer, the number of interrupted customers is estimated for
each interruption as the highest number of customers interrupted at any time during the
whole interruption duration.
7.3.4 DG Model
generating units. The FOR and repair time of DG units are assumed to be 0.01 and 44h,
respectively [109]. It should be noted that, in the case of active failure into the system,
the DG should be disconnected immediately. Once the faulty sections are isolated, the
DG supply the healthy parts of the system via operating in island mode.
7.3.5 PDM
Two different customer interruption cost models are discussed in the Chapter 6. Based
on the presented results the PDM cost model provides an opportunity to have a more
realistic assessment for the reliability of distribution systems. Therefore, in this chapter
the proposed PDM cost model based on the CCNN is considered in the analysis.
In Fig. 7.3, a set of possible locations to install sectionalizing switches and protective
devices is shown. The total cost of reliability (TCR) is formulated as an explicit nonlinear
function of decision variables indicating the installation of sectionalizing switches and
protective devices on the sections of a radial distribution network. The binary decision
variables are defined as follows:
1 if a sectionalizing switch is installed on
Xsf r = location s of feeder f r,
0 otherwise.
1 if a fuse is installed on location f of
Yf f r = feeder f r,
0 otherwise.
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 171
Sub1 Sub2
CB
: Circuit Breaker
LP1 : Tie Switch
LP3 : Sectionalizing Switch
LP2
: Fuse
LP3n
Figure 7.3: Possible locations for sectionalizing switches and protective devices in a
radial distribution network.
1 if a CB is installed on location c of
Zcf r = feeder f r,
0 otherwise.
T C = U CR + CIC (7.10)
where CIC represents the customer interruption cost and U CR is defined as the utilities
cost of reliability including costs of sectionalizing switches and protective devices.
U CR = SC + F C + CBC (7.11)
where SC is defined as the sectionalizing switch cost including capital investment cost
(CC), installation cost (IC) and maintenance cost (MC) as follows:
Nf r Ns
X X
SC = (CCs + ICs ) × Xsf r
f r=1 s=1
(7.12)
Nf r Ns
T X
X X
+ M Cs,t × Xsf r
t=1 f r=1 s=1
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 172
Nf r Nf
X X
FC = (CCf + ICf ) × Yf f r
f r=1 f =1
(7.13)
Nf r Nf
T X
X X
+ M Cf,t × Yf f r
t=1 f r=1 f =1
Nf r Nc
X X
CBC = (CCc + ICc ) × Zcf r
f r=1 c=1
(7.14)
Nf r Nc
T X
X X
+ M Cc,t × Zcf r
t=1 f r=1 c=1
The CIC consists of ECOST and the interruption cost due to temporary faults (ICT):
Nf r T Nq NLP NCT
X XX X X
CIC = (ECOSTijtkf r +
f r=1 t=1 i=1 j=1 k=1 (7.15)
The objective of the proposed formulation is to accurately model the sequence of events
after a contingency in the network. It is achieved by minimizing the total cost of reliability
in terms of customer outage cost in conjunction with sectionalizing switch and protective
device capital investment, installation, and annual operation and maintenance costs. Also,
taking into consideration the load increase rate during a time horizon under study, the
average load of load points is multiplied by (1 + γ)t−1 [23].
4- All faults are classified as the first-order failures [7]. The first-order events are
evaluated using directly the concepts of (7.1), (7.2) and (7.3), in which the values λj and
rj are the appropriate failure rate and outage duration respectively of the transient or
temporary outage. In the case of transient failures (automatic restoration) the values of
rj , may be negligible, in which case the contribution of such failures to the annual outage
time can be neglected. This may not be valid in the case of temporary failures, since the
outage time may be several hours if detection of the cause of failure is difficult and the
site of failure is in a remote rural area.
7.4.2 Constraints
This section presents economic and technical constraints that are incorporated to the
proposed MINLP model. The sectionalizing switches and CBs are comparatively expensive
devices. Hence, increasing the number of installed switches and protective devices in
distribution network will increase the UCR cost dramatically. Thus, the following economic
constraints are defined to limit the number of available switches and protective devices for
installation in the case of budget inadequacy:
Nf r Ns
X X
Xsf r ≤ Nas (7.17)
f r=1 s=1
Nf r Nf
X X
Yf f r ≤ Naf (7.18)
f r=1 f =1
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 174
Nf r Nc
X X
Zcf r ≤ Nac (7.19)
f r=1 c=1
The technical constraints are related to the network topology and utilities practices. In
fact, in the case of installing a mandatory device in a particular section, the corresponding
binary variable should be set to one. The following constraints are defined to restrict the
continuous decision variable CDFijktf r (rij ) based on the number and location of switches
and protective devices considering the switching time of sectionalizing devices, actuation
time of fuses, and repair time of faulted equipment.
Switching
CDFijktf r (rij ) ≥ CDFijktf r × (1 − Yf f r ) × Zcf r (7.20)
sj
Repair
X
CDFijktf r (rij ) ≥ [CDFijktf r × (1 − Xsf r )] × (1 − Yf f r ) × Zcf r (7.21)
s=si
Switching Repair
CDFijktf r and CDFijktf r are the fixed values and present CDF values corresponding
The first test system used in this chapter is the radial distribution system at Bus 6 of
RBTS test system which is described in [109] and presented in Fig. 7.4. The required
data, such as component failure rates, average and maximum load, number and type of
customers on feeder, etc., which are necessary to carry out the reliability studies are given
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 175
33kV
CB1
Switch
L35
Sub-Island 1 Load Point (LP)
LP18
L36 Sub-Island zone
LP19
L37
LP20
L38
LP21
L39
LP22
L40 L41
LP23
L43 L42
LP24
SW1 L44
LP31 L53 SW2
L54 L45
LP25
LP32 L55
L46
L56 L47
LP26
LP33
L48
L57
LP27
LP34 Sub-Island 3
L58 L49 SW4
L59 LP36
LP35
SW3
L60
L62 L63 L64
DGRES1
L61
Sub-Island 2 L50
DGRES3
LP28
L51
LP29 Sub-Island 5
LP30
DGRES2
Sub-Island 4
in [150]. As can be seen in Fig. 7.4, this system has four manual switches. The system
indices such as SAIFI, SAIDI and ENS values for feeder without automatic switches were
calculated and are listed in [150].
The system is divided into five sub-islands based on the location of the switches. If any
fault occurs within a sub-island, the upstream tracing method is used to find a switch or
fuse contained in this sub-island to operate. In other words, after the switching system is
installed, a fault will only affect the customers in its sub-island unless the switching system
fails or the adjacent feeders cannot pick up the transferred loads. In addition, all manual
switches are assumed to take 0.5 hour on average to act, and all automatic switches are
assumed to switch after 1 minute.
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 176
The effect of replacing CBs instead of manual switches is investigated in this section.
We changed the type of switches in this system by substituting CB in order to enhance
reliability indices of load points and of the system. For instance, when the manual switch
is installed at the start section of Sub-Island 4, the customers in this sub-island are farthest
from the main breaker, thus the failure rate and outage time for these customers are higher
than other customers. Reliability of the system will increase when CBs are added at the
start point of each sub-island instead of the manual switches. With CBs, the permanent
faults in sub-islands do not affect all parts of the system and time to restore the healthy
part of the system will decrease. As expected, the installation of the CBs enhanced the
reliability of the distribution feeder and this significant improvement is shown in Table 7.2-
Table 7.8.
The reliability index of each service sub-island defined by the boundary switches is
derived to solve the expected energy not served due to fault contingency. As CB switch
is a device that isolates a faulted part from the system so that the healthy part can still
be electrically supplied and the interruption duration and energy not supplied are mini-
mized by using the CB. Therefore, the switch type plays an important role in distribution
networks, where switches can be remotely activated.
It can be seen from the results that by substituting the CB at the start of the each
sub-island there is no improvement on the reliability indices of that sub-island but a major
effect happened on the reliability indices of the remaining sub-island and the entire system.
On the other hand, the CB contributed in preventing fault propagation to the healthy part
of the system. For instance, consider a fault in Sub-Island 2 of the test system. In this
case, when the fault occurs in Sub-Island 2, the main CB will operate and then the CB
at the start part of the sub-island operates to isolate the faulty sub-island. Thus the
reliability indices of this sub-island do not change although the reliability indices of the
other sub-islands are improved dramatically. The best performance of this strategy is when
all switches are substituted by CBs as the results depicted in Table 7.8. It can be observed
that the LPIC index has a higher improvement for the industrial customers compared to
the residential customers as they have a larger SCDF. It means that the approach is more
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 177
effective for the customers with higher SCDF. The overall improvement in ECOST was
significant in this case (around 22.7%), which represents the performance of this approach.
Meanwhile, there is reasonable improvement in the reliability indices for load points in each
sub-island especially LPENS. These results show that the type and location of switches
do have a great impact on reliability indices, which implies that the optimal placement
of switches is necessary to achieve further improvement on the reliability of distribution
systems.
Reliability of the system has increased when CBs are added at the start point of each
sub-island. With CBs, the permanent faults in sub-islands do not affect all part of the
system, and the time to restore the healthy parts of the system is decreased. Also, DG can
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 178
supply the load cut off from the main supply in the case of a fault in the zones. The best
scenario is replacing all disconnect switches by CBs. But, in the case of capital expenditure
limitations, substituting of Switches sw1, sw2 is more suitable when compared with sw3,
sw4.
occur. Existence of DG units in a distribution system will affect the restoration time of
load points and eventually reliability indices. In this network, there are four locations for
possible installation of DG units. The aim is to determine the best location and number
of DG units in order to improve the reliability indices of the system.
Case A:
In this case, the impact of DG location on system indices is assessed. The results
are shown in Table 7.9. This Table shows DG penetration based on location meliorates
reliability indices of the test system for the comparative studies. In the case of ECOST,
adding the DG unit at Sub-Island 3 and 4 produces notably better results than placing it
at Sub-Island 1. This is due to the fact that, in the test system, the DG unit can access
more customers than if just the main supply is used. When the DG units are placed close
to the main supply, not much additional benefit is expected to be achieved as corroborated
by the results. Thus, the ECOST is decreased when the DG units placed further away
from the main supply.
The ENS improves by about 10% when a DG unit is installed at the end point of
the Sub-Island 1. In the case of placing it at the end point of the Sub-Island 3, the ENS
improves by around 27%. Finally, in the case of Sub-Island 4, the ENS improves by around
31.5%.
In terms of reliability improvement, the best location for adding the DG units is at the
end of each sub-island. Once the failed sub-island is isolated, the downstream load points
can be supplied by the DG unit and the upstream load points can be served by the main
supply. Results show that indices such as ECOST and ENS are remarkably sensitive to
the location of DG units. Consequently, the location of DG units in a distribution system
is very important and the best location for DG units can be chosen from the results.
Case B:
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 181
In this case, the effect of introducing more DG units in the system is assessed. The
results are presented in Table 7.10. It is important to say that, introducing the first DG
unit remarkably enhanced the reliability indices of the system. According to the results,
installing several DG units can improve the reliability indices as compared to installing
just one. For instance, results show that the three DG units can decrease the ECOST
index by around 40%. It should be noted that this improvement depends on the location
of the DG units, type of customers and load size. However, the indices improve the most
when the number of DG units increases. This is because if one DG unit fails, the other
DG units can still supply the customers in the system. The optimum number of DG units
can be found for each location by considering financial factors and suitable index level.
The second test system is the radial distribution network at Bus 4 of the RBTS, consisting
of 38 load points, 51 possible sectionalizing switches, 38 fuses and 7 CB locations, respec-
tively [109]. The modified test network without TS is depicted in Fig. 7.5. The required
reliability data, such as customer data, component failure data, average and maximum load
at each load point are presented in [109]. The related fix costs associated with sectional-
izing switches and protective devices are presented in Table 7.11 [114]. The maintenance
cost of a switch device is assumed to be 2% of the capital investment cost. The life ex-
pectancy of the switching device is assumed to be 15 years [23]. The load growth rate of
the system is assumed to be 3%. Furthermore, the CDF data are elicited from Table 3.3.
The proposed formulation has been implemented in the GAMS environment and is solved
by using the MINLP solver BARON on an Intel(R) Core(TM) i5-2500 CPU@3.30GHz
with 8 GBs of RAM.
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 182
Device Cost($)
CB 6000
Automatic Switch 4700
Manual Switch 2500
Fuse 1500
LP14 LP17
LP11 LP12 LP13 LP15 LP16
Sectionalizing
switch 25 30
CB 20 22 24 27 29
CB3 19 21 23 26 28
Fuse F3
LP8 LP9 LP10
14 16 18
CB2 13 15 17
F2
1 3 5 7 10
F1
CB1
2 4 6 9 12
Substation at Bus 4
8 11
57 59 62 64 67
61 66
51 53 55
45 47 49
CB4 31 LP26 33 36
LP27
39
LP28
41
F4
32 35 38 40 43
34 37 42
Figure 7.5: Modified distribution test system (with 38 load points (LP), 51 sectionalizing
switches, 38 fuses and 7 CBs).
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 183
In order to apply the proposed formulation on the test network, different case studies
are performed. In the Case 1, CBs, sectionalizing switches, and fuses are installed in all
possible locations, which is the base case. The TC value is 92.365 (k$/yr). This case might
be good for the customers because of the highest reliability level, but is not a desired
option for the utilities due to the higher cost. Thus, the optimal number of protective
devices is a trade-off challenge for the utilities to keep the reliability of the network at
the reasonable levels while meeting cost constrains. In the Case 2, the optimal number
and location of manual sectionalizing switches and protective devices are investigated.
The value of TC is 70.676 (k$/yr) which is 23.5% less than for the Case 1. The results
obtained by the proposed formulation suggest installing seven CBs and show that the best
locations to install CBs are the start section of the main feeders (CB1-CB7), confirming
the performance of the proposed formulation. Note that the AAM cost model is used to
evaluate the above cases.
To have a more accurate reliability analysis, in the Case 3, the optimal number and
location of manual sectionalizing switches and protective devices are evaluated by consid-
ering the PDM cost model. The comparison between the Cases 2 and 3 is presented in
Table 7.12. As can be seen, the higher CDF is made by the PDM result in the higher
value for TC. Also the number of switches and fuses are increased due to the higher CDF.
However, despite these higher values, the PDM cost model can provide a more realistic
reliability analysis [137]. The optimal number of fuses in the Case 3 is 38, and the pro-
posed optimization formulation suggests putting fuses in all possible locations because of
their lower cost compared to the cost of customer outage.
The reliability of power distribution networks can be greatly enhanced by using au-
tomatic protective devices. Remotely controlled and automated restoration service can
avoid the required execution of manual switching schedule and is bound to bring about
remarkable levels of system reliability while reducing interruption cost. Therefore, in the
Case 4, the reliability worth of utilizing DAS in terms of automatic sectionalizing switches
(RWDAS) based on the TC is investigated. DAS in terms of automatic and remotely con-
trolled sectionalizing switches installation can provide benefits to the distribution utilities.
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 184
These benefits can be quantified in terms of reduction on the customer outage duration and
number of affected customers during a fault by fast restoration of power to the unfaulted
parts of the network. To do this, as mentioned before, the faulted part of the network
needs to be isolated by remote and automatic sectionalizing switches and the unfaulted
parts will be supplied by the main substation or alternative routes. The RWDAS can be
calculated by the following equation:
where T CDAS and T C are the network TC with and without DAS, respectively. Note
that, the PDM cost model is used in this case.
The value of RWDAS, when comparing with Case 3, is 12 351 (k$/yr) which is a major
benefit for distribution utilities. It can be evidenced that automatic switches can reduce
the objective functions (TC and ECOST) even if a higher number of installed sectionalizing
switches is proposed. In fact, the higher number of installed automatic switches the greater
reduction on the outage time of customers, consequently improving the objective functions
and continuity of supply. By inspecting the results, it is important to replace the manual
switch devices with automatic devices to reduce the restoration time which is a crucial
measure to decrease the customer interruption costs. Furthermore, the proposed MINLP
formulation suggests to place 15 switches, 38 fuses and 7 CBs in the test network for the
Case 4.
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 185
Location and capacity of DG units are important to improve the quality of service in
distribution networks [151], [152]. However, such improvement depends on DG systems
operating in islanding mode, which is adopted in this dissertation. Island formation oc-
curs when a part of distribution network that includes one or more DG units becomes
electrically isolated from the rest of the network and continues to be energized by the DG.
An island is successful when DG units do not have faults and their generation capacity
is higher or at least equal to the load demand within the island. In this chapter, it is
assumed that the load points in the island will be supplied continuously when the island
is forming.
NLP f
X
PIP = ( (PGDG × PGLPj )) × (1 − Pf ) (7.23)
j=1
where PGDG represents the cumulative probability of DG units generating power equal
to or greater than a certain level, PGLP denotes the probability of a load point having a
certain value, and Pf is the forced outage rate of the DG units.
The major players in renewable energy generation are solar photovoltaic, wind, biomass
and fuel cell [153]. The increase in fuel price has prompted distribution network operators
to invest in renewable energy sources [154]. In this chapter, two DG technology options,
namely wind turbines (consisting of two wind turbines each of 1 MW rated output) and a
diesel generator (with a rating of 2 MW) are considered. The probability function for the
wind turbines output power and network loads are extracted from [122].
The DG units that have the ability to supply the loads in the feeder are introduced
into the modified test network. As the addition of DG units on the feeders supplying the
customers with higher CDF and heavier loads is more effective to reduce the TC of the
network, in the Case 5, Feeder 4 (F4) is selected for adding a DG unit into the modified
Chapter 7. Distribution System Protective Devices Placement Optimization Considering
DAS and PDM 186
test network. The optimal number and location of switches and protective devices of Case
5 are presented in Table 7.13. Note that, the DAS and PDM cost models are considered
in this case.
The results indicate that adding a DG unit at the end of Feeder 4 can decrease the TC
by about $17 675 (k$/yr) compared to the Case 4. This is due to the fact that this feeder
(F4) has two commercial customers with relatively high CDFs and has the highest load
level of the entire studied network. As can be seen, 17 switches are selected to be placed
in the network by the proposed formulation. In fact, adding DG can reduce the objective
function even if a higher number of installed switches is proposed. This is due to the fact
that the higher number of installed switches, the greater reduction on the outage time of
customers.
7.7 Conclusion
This chapter has presented a formulation to identify the optimal number, types and loca-
tions of protective devices and sectionalizing switches in distribution networks. A MINLP
optimization has been implemented to assess the effects of DAS in improving the reliability
indices of a distribution network. The objective was to minimize the TC while installing
the minimum number of protective devices and switches. Furthermore, to have a more
accurate model for the reliability assessment, which can provide better solutions to meet
the utility practices, the PDM interruption cost model is considered. The effectiveness
DAS and PDM
Table 7.13: Impact of DG’s Number on Optimal Number and Location of Sectionalizing Switches
of the proposed formulation has been validated by means of case studies on Bus 4 of the
modified RBTS. Numerical results and sensitivity analyses have confirmed the efficient
performance of this formulation. Also, in the case of adding DG units, the effects of
DAS and the PDM on the optimal protective and switching devices placement have been
investigated and discussed.
Chapter 8
This thesis has reported the reliability analysis of the distribution networks with DG units
in terms of protective devices placement. The two MILP and MINLP models are have
been proposed to identify optimum type, number and location of protective devices. A
value-based reliability optimization formulation by considering the customer outage and
protective devices costs is derived from the proposed model. Also, a new islanding scheme
is proposed to benefit the networks in terms of reliability improvement by introducing DG
units. To have a more accurate reliability estimation, a PDM cost model based on the
CCNN is proposed. Furthermore, to ensure the effectiveness of the proposed techniques,
many sensitivity analysis have been performed on the test distribution network.
8.1 Conclusions
Chapter 2 introduces the basic concept of distribution networks reliability and different
methods to assess the reliability. The reliability evaluation of the three traditional hi-
erarchical levels of a power system was reported. A wide range of methods have been
used to evaluate the reliability of power systems. These methods can be divided into two
categories of simulation and analytical approaches. Analytical methods often determine
the system topology and input values and then create a mathematical model of the power
189
Chapter 8. Conclusions and Future Work 190
system to evaluate the reliability. Simulation approaches such as Monte Carlo can de-
termine the indices by simulating the probabilistic and random behavior of the systems.
Therefore, the problem is treated as a series of experiments instead of considering the
analytical models of the systems. The main weakness of simulation methods is that to
obtain an acceptable level of the accuracy of the performance indices a huge amount of
experiments need to run. Due to the long computational time, simulation approaches are
not as popular as analytical approaches. The reinforcements to improve the reliability of
distribution networks are also presented in this chapter.
One of the challenges for the utilities is to provide an acceptable level of reliability which
can meet the customer demands while minimizing the associated costs. An effective way is
to identify devices and structures to benefit customers and utilities with the lowest possible
costs. The optimum number and location of these devices are important considerations in
power distribution system planning. Sectionalizing switches are the solution that permits
the system to be isolated following a fault and provides power restoration. The main
challenge is that increasing levels of reliability requires additional switches and greater
capital and operational expenditure. It is therefore important to optimization the number
and location of sectionalizing switches to find a balance between reliability improvement
and cost. Regarding the structure, introducing DG units can potentially improve network
reliability, as it can reduce interruption duration and restoration time of the customers
in the distribution networks. Therefore, the problem of selecting optimum number and
suitable locations of sectionalizing switches can be more complicated when DG units are
added in the distribution network and still is an important challenge for the utilities and
network operators.
at the end of feeder with higher load levels and consisting of commercial customers can
significantly reduces the total cost of reliability even if a higher number of sectionalizing
switches is proposed to install. By increasing the number of DG units in the system, the
proposed formulation suggested to install more switches. In fact, installing more switches
provides an opportunity to supply more customers and decreases the objective function
(TC) related to the solution.
The analysis shows that the sectionalizing switches are not the only devices that can
improve the reliability. Different types of protective devices can be used in distribution net-
works to meet the desire reliability from the customer and utility point of view. Therefore,
the optimized allocation of protective devices and sectionalizing switches in distribution
networks can improve the quality of supply and reliability indices.
A new MINLP formulation to determine the type, number and location of protective
devices and sectionalizing switches was proposed in Chapter 5. Based on the technical and
economic constraints, the proposed formulation is used to minimize the cost of reliability
while the optimum number and location of protective devices have been identified. A
value-based reliability methodology considering the cost of customer outage and protective
devices has been implemented to minimize the reliability indices of a distribution network.
To meet the utility practices and improve the reliability, the main goal of the proposed
formulation was developing a more accurate model for the fault clearance, separation and
power restoration procedure. Furthermore, the effects of TS, pre-installed devices, budget
limitation and the importance of customer type on the optimal protective and switching
devices placement were investigated through sensitivity analysis. The proposed method
has been tested on Bus 4 of the modified RBTS. A formulation to calculate the annual
maintenance cost of a fuse was also presented in this chapter. The suggested number and
location of protective devices have confirmed the efficient performance of the proposed
formulation.
The customer interruption cost data might have dispersed nature and was discussed
in Chapter 6. A CCNN has been developed to construct the two cost models. The first
model was called the AAM in which the dispersed nature of cost data was not included
in this model. The second model was the PDM cost model in which the dispersed nature
Chapter 8. Conclusions and Future Work 193
of cost data was considered in this model. These two cost models were extracted for
the residential and industrial customers based on the survey data. An analytical method
has been used to evaluate the reliability indices and calculate the interruption cost of
individual load point and system. A combined three-state model was considered for the
component in the system. The loads in the system were divided into ten levels using a
clustering technique. The benefits of introducing DG units, which can operate in island
mode, in distribution networks have been reported. DG units were modeled by a two-state
model. The reliability worth of introducing DG units in the test distribution system were
also investigated using two cost models. The results showed that the PDM cost model
provides a better representation of the monetary losses incurred by the customers due to
the power failure than the AAM. In fact, the AAM cost model resulted in underestimating
the reliability worth of network reinforcement.
This dissertation provided modeling and formulation of the protective and section-
alizing switching devices optimization problem considering a variety of factors. The
Bibliography 194
The proposed MILP and MINLP formulations in this dissertation may be expanded
to solve the problem of optimal placement of protective devices and DG units simul-
taneously.
Bibliography
[1] R. Brown, Electric Power Distribution Reliability. New York Marcel Dekker, 2002.
[2] J.-M. Sohn, S.-R. Nam, and J.-K. Park, “Value-based radial distribution system
reliability optimization,” IEEE Trans. on Power Systems, vol. 21, no. 2, pp. 941–
947, May. 2006.
[4] J. Endrenyi, Reliability Modeling in Electric Power Systems. John Wiley & Sons,
New York, NY, 1978.
[5] J.-H. Teng and C.-N. Lu, “Feeder-switch relocation for customer interruption cost
minimization,” IEEE Trans. on Power Delivery, vol. 17, no. 1, pp. 254–259, Jan.
2002.
[7] R. Billinton and R. Allan, Reliability Evaluation of Power Systems. Springer, 1996.
195
Bibliography 196
[11] S. Elsayed, R. Sarker, and D. Essam, “An improved self-adaptive differential evolu-
tion algorithm for optimization problems,” IEEE Trans. on Industrial Informatics,
vol. 9, no. 1, pp. 89–99, Feb 2013.
[13] G. Celli and F. Pilo, “Optimal sectionalizing switches allocation in distribution net-
works,” IEEE Trans. on Power Delivery, vol. 14, no. 3, pp. 1167–1172, Jul. 1999.
[14] “IEEE Standard for Interconnecting Distributed Resources into Electric Power Sys-
tems,” IEEE Std. 1547-2003, p. 16, June 2003.
[16] Y. Atwa and E. El-Saadany, “Reliability evaluation for distribution system with
renewable distributed generation during islanded mode of operation,” IEEE Trans.
Power Systems, vol. 24, no. 2, pp. 572–581, May 2009.
[17] S. Chen, Y. Wu, and B. L. Luk, “Combined genetic algorithm optimization and
regularized orthogonal least squares learning for radial basis function networks,”
IEEE Trans. Neural Networks,, vol. 10, no. 5, pp. 1239–1243, 1999.
[25] “Glossary of terms used in NERC reliability standards,” NERC, Tech. Rep., 2014.
[26] M. S. Masoum and E. Fuchs, Power Quality in Power Systems and Electrical Ma-
chines. Academic Press, Jul. 2015.
power system reliability criteria and indices-trends and future needs,” IEEE Trans.
on Power Systems, vol. 9, no. 1, pp. 181–190, 1994.
[30] ——, “Bulk power system reliability assessment-why and how? part ii: How?” IEEE
Trans. on Power Apparatus and Systems, vol. PAS-101, no. 9, pp. 3446–3456, 1982.
[33] R. Billinton and L. Goel, “Overall adequacy assessment of an electric power system,”
IEE Proceedings Generation, Transmission and Distribution, vol. 139, no. 1, pp. 57–
63, 1992.
[36] K. Hagkwen and C. Singh, “Reliability modeling and simulation in power systems
with aging characteristics,” IEEE Trans. on Power Systems, vol. 25, no. 1, pp. 21–28,
2010.
[37] K. Hagkwen and C.Singh, “Power system reliability modeling with aging using thin-
ning algorithm,” in IEEE Bucharest PowerTech, 2009, pp. 1–6.
Bibliography 199
[38] K. Hagkwen and C. Singh, “Composite power system reliability modeling and evalu-
ation considering aging components,” in International Conference on Electrical and
Electronics Engineering, 2009, pp. 14–18.
[40] P. S. Hale, R. G. Arno, and D. O. Koval, “Analysis techniques for electrical and me-
chanical power systems,” in IEEE ndustrial and Commercial Power Systems Tech-
nical Conference, 2001, pp. 61–65.
[44] W. Li, Risk Assessment Of Power Systems: Models, Methods, and Applications.
Wiley-IEEE Press, 2004.
[45] W. E. Vesely, “Fault tree handbook: Nuclear regulatory commission,” Tech. Rep.,
1987.
[50] S. T. Lee, “Probabilistic reliability assessment for transmission planning and oper-
ation including cascading outages,” in IEEE/PES Power Systems Conference and
Exposition, 2009, pp. 1–8.
[52] K. Alvehag and L. Soder, “A reliability model for distribution systems incorporating
seasonal variations in severe weather,” IEEE Trans. on Power Delivery, vol. 26, no. 2,
pp. 910–919, 2011.
[55] Z. S. Roldan, C. Shan Heng, and R. J. Lee, “Assessing the value of distribution in-
frastructure replacement,” in IEEE/PES Power Systems Conference and Exposition,
2009, pp. 1–7.
[56] T. Wang and Y. Wu, “The reliability evaluating method considering component
aging for distribution network,” Energy Procedia, vol. 16, Part C, no. 0, pp. 1613–
1618, 2012.
[58] I. Dobson, “Where is the edge for cascading failure?: challenges and opportunities
for quantifying blackout risk,” in IEEE Power Engineering Society General Meeting,
2005, pp. 1–8.
[59] L. Wenyuan and R. Billinton, “Common cause outage models in power system re-
liability evaluation,” IEEE Trans. on Power Systems, vol. 18, no. 2, pp. 966–968,
2003.
[60] K. Yamashita, L. Juan, Z. Pei, and L. Chen-Ching, “Analysis and control of major
blackout events,” in Power Systems Conference and Exposition, PSCE, 2009, pp.
1–4.
[62] M. Ding, F. Cui, J. Wang, Y. Zheng, and S. Li, “Reliability evaluation of distribu-
tion system with wind farms based on network-equivalent method,” in 2nd IEEE
International Symposium on Power Electronics for Distributed Generation Systems
(PEDG), 2010, pp. 958–963.
[65] R. Billinton and W. Li, “A system state transition sampling method for composite
system reliability evaluation,” IEEE Trans. on Power Systems, vol. 8, no. 3, pp.
761–770, 1993.
[66] A. Sankarakrishnan and R. Billinton, “Sequential monte carlo simulation for com-
posite power system reliability analysis with time varying loads,” IEEE Trans. on
Power Systems, vol. 10, no. 3, pp. 1540–1545, 1995.
Bibliography 202
[67] R. Billinton and W. Peng, “Teaching distribution system reliability evaluation using
monte carlo simulation,” IEEE Trans. on Power Systems, vol. 14, no. 2, pp. 397–403,
1999.
[69] M. Yiming and K. N. Miu, “Switch placement to improve system reliability for radial
distribution systems with distributed generation,” IEEE Trans. on Power Systems,
vol. 18, no. 4, pp. 1346–1352, 2003.
[73] D. Haughton and G. Heydt, “Smart distribution system design: Automatic reconfig-
uration for improved reliability,” in IEEE Power and Energy Society General Meet-
ing, 2010, pp. 1 –8.
[75] C. Singh and Y. Kim, “An efficient technique for reliability analysis of power systems
including time dependent sources,” IEEE Trans. on Power Systems, vol. 3, no. 3,
pp. 1090–1096, 1988.
Bibliography 203
[76] P. Wang and R. Billinton, “Time-sequential simulation technique for rural distribu-
tion system reliability cost/worth evaluation including wind generation as alternative
supply,” IEE Proceedings Generation, Transmission and Distribution, vol. 148, no. 4,
pp. 355–360, 2001.
[77] S. X. Wang, Z. Wei, and Y. Y. Chen, “Distribution system reliability evaluation con-
sidering dg impacts,” in Third International Conference on Electric Utility Deregu-
lation and Restructuring and Power Technologies, pp. 2603–2607.
[83] H. Karimi, A. Yazdani, and R. Iravani, “Negative-sequence current injection for fast
islanding detection of a distributed resource unit,” IEEE Trans. Power Electronics,
vol. 23, no. 1, pp. 298–307, 2008.
[84] W.-Y. Chang, “A hybrid islanding detection method for distributed synchronous
generators,” in Proc. International Power Electronics Conference, 2010, pp. 1326–
1330.
Bibliography 204
[85] P. Mahat, Z. Chen, and B. Bak-Jensen, “Control and operation of distributed gener-
ation in distribution systems,” Electric Power Systems Research, vol. 81, no. 2, pp.
495–502, 2011.
[88] P. Fuangfoo, W.-J. Lee, and M.-T. Kuo, “Impact study on intentional islanding of
distributed generation connected to a radial subtransmission system in thailand’s
electric power system,” IEEE Trans. Industry Applications, vol. 43, no. 6, pp. 1491–
1498, Nov 2007.
[92] F. Viawan, D. Karlsson, A. Sannino, and J. Daalder, “Protection scheme for meshed
distribution systems with high penetration of distributed generation,” in Proc. Power
Systems Conference, 2006, pp. 99–104.
[93] S. Brahma and A. Girgis, “Development of adaptive protection scheme for distribu-
tion systems with high penetration of distributed generation,” IEEE Trans. Power
Delivery, vol. 19, no. 1, pp. 56–63, Jan 2004.
Bibliography 205
[94] D. MacKay, “Bayesian interpolation,” in Neural Copmuter, vol. 4, no. 3, 1992, pp.
415–447.
[96] W. fa Wang, Z. lin Li, and Y. Ma, “Application of backtracking algorithm in college
dormitory assignment management,” in Proc. 2nd IEEE International Conference
on Computer Science and Information Technology,ICCSIT, 2009, pp. 272–274.
[97] Y. Chen, Z. Wang, and X. Liu, “Automated point feature label placement using
backtracking algorithm with an adjacent graph,” in Proc. 18th International Con-
ference on Geoinformatics, 2010, pp. 1–5.
[98] F. Viawan, D. Karlsson, A. Sannino, and J. Daalder, “Protection scheme for meshed
distribution systems with high penetration of distributed generation,” in Proc. Power
Systems Conference: Advanced Metering, Protection, Control, Communication, and
Distributed Resources, 2006, pp. 99–104.
[102] ABB. Rel 670, 512 numerical transmission line protection system. [Online].
Available: http://www.abb.com
Bibliography 206
[104] E. Cardozo and S. N. Talukdar, “A distributed expert system for fault diagnosis,”
IEEE Trans. Power Systems, vol. 3, no. 2, pp. 641–646, May 1988.
[105] A. Girgis and M. B. Johns, “A hybrid expert system for faulted section identification,
fault type classification and selection of fault location algorithms,” IEEE Trans.
Power Delivery, vol. 4, no. 2, pp. 978–985, Apr 1989.
[106] A. Wiszniewski, “Accurate fault impedance locating algorithm,” IEE Proce. Gen-
eration, Transmission and Distribution,, vol. 130, no. 6, pp. 311–314, November
1983.
[107] I. Balaguer, Q. Lei, S. Yang, U. Supatti, and F. Z. Peng, “Control for grid-connected
and intentional islanding operations of distributed power generation,” IEEE Trans.
on Industrial Electronics, vol. 58, no. 1, pp. 147–157, Jan 2011.
[108] K. Zou, A. Agalgaonkar, K. Muttaqi, and S. Perera, “An analytical approach for re-
liability evaluation of distribution systems containing dispatchable and nondispatch-
able renewable dg units,” IEEE Trans. on Smart Grid, vol. 5, no. 6, pp. 2657–2665,
Nov 2014.
[109] R. Allan, R. Billinton, I. Sjarief, L. Goel, and K. So, “A reliability test system
for educational purposes-basic distribution system data and results,” IEEE Trans.
Power Systems, vol. 6, pp. 813 –820, May. 1991.
[110] R. Billinton and R. Karki, “Maintaining supply reliability of small isolated power
systems using renewable energy,” IEE Proceedings Generation, Transmission and
Distribution, vol. 148, no. 6, pp. 530–534, Nov 2001.
[111] “IEEE Recommended Practice for Protection and Coordination of Industrial and
Commercial Power Systems,” IEEE Std 242-2001 (Revision of IEEE Std 242-1986),
pp. 1–710, 2001.
Bibliography 207
[112] N. Rezaei and M.-R. Haghifam, “Protection scheme for a distribution system with
distributed generation using neural networks,” International Journal of Electrical
Power and Energy Systems, vol. 30, no. 4, pp. 235 – 241, 2008.
[113] S. Javadian, M.-R. Haghifam, and N. Rezaei, “A fault location and protection scheme
for distribution systems in presence of dg using mlp neural networks,” in Proc. IEEE
PES, July 2009, pp. 1–8.
[116] M. Jafari, H. Monsef, and S. Moghadam, “Optimum number, placement and capacity
of dgs and reclosers using analysis hierarchical process and genetic algorithm,” in
Proc. 10th International Conference on Environment and Electrical Engineering,
2011, pp. 1–4.
[118] C.-S. Chen, C.-H. Lin, H.-J. Chuang, C.-S. Li, M.-Y. Huang, and C.-W. Huang,
“Optimal placement of line switches for distribution automation systems using im-
mune algorithm,” IEEE Trans. on Power Systems, vol. 21, no. 3, pp. 1209–1217,
Aug. 2006.
[119] G. Chen, “A novel qea-based optimum switch placement method for improving cus-
tomer service reliability,” in Proc. Third International Conference on Electric Utility
Deregulation and Restructuring and Power Technologies, 2008, pp. 620–623.
[121] H. Falaghi, M. Haghifam, and C. Singh, “Ant colony optimization-based method for
placement of sectionalizing switches in distribution networks using a fuzzy multiob-
jective approach,” IEEE Trans. on Power Delivery, vol. 24, no. 1, pp. 268–276, Jan.
2009.
[122] Y. Atwa and E. El-Saadany, “Reliability evaluation for distribution system with
renewable distributed generation during islanded mode of operation,” IEEE Trans.
on Power Systems, vol. 24, no. 2, pp. 572–581, 2009.
[123] I. Lim, T. Sidhu, M. Choi, S. Lee, and B. Ha, “An optimal composition and place-
ment of automatic switches in das,” IEEE Trans. on Power Delivery, vol. 28, no. 3,
pp. 1474–1482, Jul. 2013.
[125] F. Soudi and K. Tomsovic, “Optimized distribution protection using binary pro-
gramming,” IEEE Trans. on Power Delivery, vol. 13, no. 1, pp. 218–224, Jan. 1998.
[126] G. Ferreira and A. Bretas, “A nonlinear binary programming model for electric dis-
tribution systems reliability optimization,” International Journal of Electrical Power
and Energy Systems, vol. 43, no. 1, pp. 384–392, 2012.
[128] “IEEE Guide for Electric Power Distribution Reliability Indices,” IEEE Std. 1366-
2012 (Revision of IEEE Std. 1366-2003), pp. 1–43, May 2012.
[129] N. Ding, Y. Besanger, F. Wurtz, and G. Antoine, “Individual nonparametric load es-
timation model for power distribution network planning,” IEEE Trans. on Industrial
Informatics, vol. 9, no. 3, pp. 1578–1587, Aug 2013.
Bibliography 209
[133] T. Kohonen, “The self-organizing map,” Proc. IEEE, vol. 78, no. 9, pp. 1464–1480,
Sep. 1990.
[134] G. Huang, S. Song, and C. Wu, “Orthogonal least squares algorithm for training
cascade neural networks,” IEEE Trans. Circuits and Systems, vol. 59, no. 11, pp.
2629–2637, Nov. 2012.
[135] B. Wilamowski, “Neural network architectures and learning algorithms,” IEEE Ind.
Electron. Mag, vol. 3, no. 4, pp. 56–63, Dec. 2009.
[137] R. Billinton and P. Wang, “Reliability worth of distribution system network rein-
forcement considering dispersed customer cost data,” IEE Proc. Gener. Transm.
Distrib., vol. 146, no. 3, pp. 318–324, May 1999.
[142] Y. Atwa and E. El-Saadany, “Reliability evaluation for distribution system with
renewable distributed generation during islanded mode of operation,” IEEE Trans.
Power Systems, vol. 24, no. 2, pp. 572–581, May 2009.
[146] I. Dzafic, R. Jabr, E. Halilovic, and B. Pal, “A sensitivity approach to model lo-
cal voltage controllers in distribution networks,” IEEE Trans. on Power Systems,
vol. 29, no. 3, pp. 1419–1428, May. 2014.
[148] O. Tortelli, E. Lourenco, A. Garcia, and B. Pal, “Fast decoupled power flow to emerg-
ing distribution systems via complex pu normalization,” IEEE Trans. on Power
Systems, vol. 30, no. 3, May. 2015.
Bibliography 211
[149] M. Ariff, B. Pal, and A. Singh, “Estimating dynamic model parameters for adaptive
protection and control in power system,” IEEE Trans. on Power Systems, vol. 30,
no. 2, pp. 829–839, Mar. 2015.
[150] R. Billinton and S. Jonnavithula, “A test system for teaching overall power system
reliability assessment,” IEEE Trans. Power Systems, vol. 11, pp. 1670–1676, 1996.
[151] R. Jabr and B. Pal, “Ordinal optimisation approach for locating and sizing of dis-
tributed generation,” IET Generation, Transmission Distribution, vol. 3, no. 8, pp.
713–723, Aug. 2009.
[153] H. Abu-Rub, A. Iqbal, S. Moin Ahmed, F. Peng, Y. Li, and G. Baoming, “Quasi-z-
source inverter-based photovoltaic generation system with maximum power tracking
control using anfis,” IEEE Trans. on Sustainable Energy, vol. 4, no. 1, pp. 11–20,
Jan. 2013.
[154] R. Jabr and B. Pal, “Intermittent wind generation in optimal power flow dispatch-
ing,” IET Generation, Transmission and Distribution., vol. 3, no. 1, pp. 66–74, Jan.
2009.